1 /*- 2 * SPDX-License-Identifier: BSD-3-Clause 3 * 4 * Copyright (c) 2007-2009 Google Inc. and Amit Singh 5 * All rights reserved. 6 * 7 * Redistribution and use in source and binary forms, with or without 8 * modification, are permitted provided that the following conditions are 9 * met: 10 * 11 * * Redistributions of source code must retain the above copyright 12 * notice, this list of conditions and the following disclaimer. 13 * * Redistributions in binary form must reproduce the above 14 * copyright notice, this list of conditions and the following disclaimer 15 * in the documentation and/or other materials provided with the 16 * distribution. 17 * * Neither the name of Google Inc. nor the names of its 18 * contributors may be used to endorse or promote products derived from 19 * this software without specific prior written permission. 20 * 21 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS 22 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT 23 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR 24 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT 25 * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, 26 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT 27 * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, 28 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY 29 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT 30 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE 31 * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 32 * 33 * Copyright (C) 2005 Csaba Henk. 34 * All rights reserved. 35 * 36 * Copyright (c) 2019 The FreeBSD Foundation 37 * 38 * Portions of this software were developed by BFF Storage Systems, LLC under 39 * sponsorship from the FreeBSD Foundation. 40 * 41 * Redistribution and use in source and binary forms, with or without 42 * modification, are permitted provided that the following conditions 43 * are met: 44 * 1. Redistributions of source code must retain the above copyright 45 * notice, this list of conditions and the following disclaimer. 46 * 2. Redistributions in binary form must reproduce the above copyright 47 * notice, this list of conditions and the following disclaimer in the 48 * documentation and/or other materials provided with the distribution. 49 * 50 * THIS SOFTWARE IS PROVIDED BY AUTHOR AND CONTRIBUTORS ``AS IS'' AND 51 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 52 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 53 * ARE DISCLAIMED. IN NO EVENT SHALL AUTHOR OR CONTRIBUTORS BE LIABLE 54 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 55 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 56 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 57 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 58 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 59 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 60 * SUCH DAMAGE. 61 */ 62 63 #include <sys/cdefs.h> 64 __FBSDID("$FreeBSD$"); 65 66 #include <sys/param.h> 67 #include <sys/module.h> 68 #include <sys/systm.h> 69 #include <sys/errno.h> 70 #include <sys/kernel.h> 71 #include <sys/conf.h> 72 #include <sys/filio.h> 73 #include <sys/uio.h> 74 #include <sys/malloc.h> 75 #include <sys/queue.h> 76 #include <sys/limits.h> 77 #include <sys/lock.h> 78 #include <sys/rwlock.h> 79 #include <sys/sx.h> 80 #include <sys/proc.h> 81 #include <sys/mount.h> 82 #include <sys/vnode.h> 83 #include <sys/namei.h> 84 #include <sys/extattr.h> 85 #include <sys/stat.h> 86 #include <sys/unistd.h> 87 #include <sys/filedesc.h> 88 #include <sys/file.h> 89 #include <sys/fcntl.h> 90 #include <sys/dirent.h> 91 #include <sys/bio.h> 92 #include <sys/buf.h> 93 #include <sys/sysctl.h> 94 #include <sys/vmmeter.h> 95 96 #include <vm/vm.h> 97 #include <vm/vm_extern.h> 98 #include <vm/pmap.h> 99 #include <vm/vm_map.h> 100 #include <vm/vm_page.h> 101 #include <vm/vm_param.h> 102 #include <vm/vm_object.h> 103 #include <vm/vm_pager.h> 104 #include <vm/vnode_pager.h> 105 #include <vm/vm_object.h> 106 107 #include "fuse.h" 108 #include "fuse_file.h" 109 #include "fuse_internal.h" 110 #include "fuse_ipc.h" 111 #include "fuse_node.h" 112 #include "fuse_io.h" 113 114 #include <sys/priv.h> 115 116 /* Maximum number of hardlinks to a single FUSE file */ 117 #define FUSE_LINK_MAX UINT32_MAX 118 119 SDT_PROVIDER_DECLARE(fusefs); 120 /* 121 * Fuse trace probe: 122 * arg0: verbosity. Higher numbers give more verbose messages 123 * arg1: Textual message 124 */ 125 SDT_PROBE_DEFINE2(fusefs, , vnops, trace, "int", "char*"); 126 127 /* vnode ops */ 128 static vop_access_t fuse_vnop_access; 129 static vop_advlock_t fuse_vnop_advlock; 130 static vop_allocate_t fuse_vnop_allocate; 131 static vop_bmap_t fuse_vnop_bmap; 132 static vop_close_t fuse_fifo_close; 133 static vop_close_t fuse_vnop_close; 134 static vop_copy_file_range_t fuse_vnop_copy_file_range; 135 static vop_create_t fuse_vnop_create; 136 static vop_deallocate_t fuse_vnop_deallocate; 137 static vop_deleteextattr_t fuse_vnop_deleteextattr; 138 static vop_fdatasync_t fuse_vnop_fdatasync; 139 static vop_fsync_t fuse_vnop_fsync; 140 static vop_getattr_t fuse_vnop_getattr; 141 static vop_getextattr_t fuse_vnop_getextattr; 142 static vop_inactive_t fuse_vnop_inactive; 143 static vop_ioctl_t fuse_vnop_ioctl; 144 static vop_link_t fuse_vnop_link; 145 static vop_listextattr_t fuse_vnop_listextattr; 146 static vop_lookup_t fuse_vnop_lookup; 147 static vop_mkdir_t fuse_vnop_mkdir; 148 static vop_mknod_t fuse_vnop_mknod; 149 static vop_open_t fuse_vnop_open; 150 static vop_pathconf_t fuse_vnop_pathconf; 151 static vop_read_t fuse_vnop_read; 152 static vop_readdir_t fuse_vnop_readdir; 153 static vop_readlink_t fuse_vnop_readlink; 154 static vop_reclaim_t fuse_vnop_reclaim; 155 static vop_remove_t fuse_vnop_remove; 156 static vop_rename_t fuse_vnop_rename; 157 static vop_rmdir_t fuse_vnop_rmdir; 158 static vop_setattr_t fuse_vnop_setattr; 159 static vop_setextattr_t fuse_vnop_setextattr; 160 static vop_strategy_t fuse_vnop_strategy; 161 static vop_symlink_t fuse_vnop_symlink; 162 static vop_write_t fuse_vnop_write; 163 static vop_getpages_t fuse_vnop_getpages; 164 static vop_print_t fuse_vnop_print; 165 static vop_vptofh_t fuse_vnop_vptofh; 166 167 struct vop_vector fuse_fifoops = { 168 .vop_default = &fifo_specops, 169 .vop_access = fuse_vnop_access, 170 .vop_close = fuse_fifo_close, 171 .vop_fsync = fuse_vnop_fsync, 172 .vop_getattr = fuse_vnop_getattr, 173 .vop_inactive = fuse_vnop_inactive, 174 .vop_pathconf = fuse_vnop_pathconf, 175 .vop_print = fuse_vnop_print, 176 .vop_read = VOP_PANIC, 177 .vop_reclaim = fuse_vnop_reclaim, 178 .vop_setattr = fuse_vnop_setattr, 179 .vop_write = VOP_PANIC, 180 .vop_vptofh = fuse_vnop_vptofh, 181 }; 182 VFS_VOP_VECTOR_REGISTER(fuse_fifoops); 183 184 struct vop_vector fuse_vnops = { 185 .vop_allocate = fuse_vnop_allocate, 186 .vop_default = &default_vnodeops, 187 .vop_access = fuse_vnop_access, 188 .vop_advlock = fuse_vnop_advlock, 189 .vop_bmap = fuse_vnop_bmap, 190 .vop_close = fuse_vnop_close, 191 .vop_copy_file_range = fuse_vnop_copy_file_range, 192 .vop_create = fuse_vnop_create, 193 .vop_deallocate = fuse_vnop_deallocate, 194 .vop_deleteextattr = fuse_vnop_deleteextattr, 195 .vop_fsync = fuse_vnop_fsync, 196 .vop_fdatasync = fuse_vnop_fdatasync, 197 .vop_getattr = fuse_vnop_getattr, 198 .vop_getextattr = fuse_vnop_getextattr, 199 .vop_inactive = fuse_vnop_inactive, 200 .vop_ioctl = fuse_vnop_ioctl, 201 .vop_link = fuse_vnop_link, 202 .vop_listextattr = fuse_vnop_listextattr, 203 .vop_lookup = fuse_vnop_lookup, 204 .vop_mkdir = fuse_vnop_mkdir, 205 .vop_mknod = fuse_vnop_mknod, 206 .vop_open = fuse_vnop_open, 207 .vop_pathconf = fuse_vnop_pathconf, 208 /* 209 * TODO: implement vop_poll after upgrading to protocol 7.21. 210 * FUSE_POLL was added in protocol 7.11, but it's kind of broken until 211 * 7.21, which adds the ability for the client to choose which poll 212 * events it wants, and for a client to deregister a file handle 213 */ 214 .vop_read = fuse_vnop_read, 215 .vop_readdir = fuse_vnop_readdir, 216 .vop_readlink = fuse_vnop_readlink, 217 .vop_reclaim = fuse_vnop_reclaim, 218 .vop_remove = fuse_vnop_remove, 219 .vop_rename = fuse_vnop_rename, 220 .vop_rmdir = fuse_vnop_rmdir, 221 .vop_setattr = fuse_vnop_setattr, 222 .vop_setextattr = fuse_vnop_setextattr, 223 .vop_strategy = fuse_vnop_strategy, 224 .vop_symlink = fuse_vnop_symlink, 225 .vop_write = fuse_vnop_write, 226 .vop_getpages = fuse_vnop_getpages, 227 .vop_print = fuse_vnop_print, 228 .vop_vptofh = fuse_vnop_vptofh, 229 }; 230 VFS_VOP_VECTOR_REGISTER(fuse_vnops); 231 232 uma_zone_t fuse_pbuf_zone; 233 234 /* Check permission for extattr operations, much like extattr_check_cred */ 235 static int 236 fuse_extattr_check_cred(struct vnode *vp, int ns, struct ucred *cred, 237 struct thread *td, accmode_t accmode) 238 { 239 struct mount *mp = vnode_mount(vp); 240 struct fuse_data *data = fuse_get_mpdata(mp); 241 int default_permissions = data->dataflags & FSESS_DEFAULT_PERMISSIONS; 242 243 /* 244 * Kernel-invoked always succeeds. 245 */ 246 if (cred == NOCRED) 247 return (0); 248 249 /* 250 * Do not allow privileged processes in jail to directly manipulate 251 * system attributes. 252 */ 253 switch (ns) { 254 case EXTATTR_NAMESPACE_SYSTEM: 255 if (default_permissions) { 256 return (priv_check_cred(cred, PRIV_VFS_EXTATTR_SYSTEM)); 257 } 258 return (0); 259 case EXTATTR_NAMESPACE_USER: 260 if (default_permissions) { 261 return (fuse_internal_access(vp, accmode, td, cred)); 262 } 263 return (0); 264 default: 265 return (EPERM); 266 } 267 } 268 269 /* Get a filehandle for a directory */ 270 static int 271 fuse_filehandle_get_dir(struct vnode *vp, struct fuse_filehandle **fufhp, 272 struct ucred *cred, pid_t pid) 273 { 274 if (fuse_filehandle_get(vp, FREAD, fufhp, cred, pid) == 0) 275 return 0; 276 return fuse_filehandle_get(vp, FEXEC, fufhp, cred, pid); 277 } 278 279 /* Send FUSE_FLUSH for this vnode */ 280 static int 281 fuse_flush(struct vnode *vp, struct ucred *cred, pid_t pid, int fflag) 282 { 283 struct fuse_flush_in *ffi; 284 struct fuse_filehandle *fufh; 285 struct fuse_dispatcher fdi; 286 struct thread *td = curthread; 287 struct mount *mp = vnode_mount(vp); 288 int err; 289 290 if (fsess_not_impl(vnode_mount(vp), FUSE_FLUSH)) 291 return 0; 292 293 err = fuse_filehandle_getrw(vp, fflag, &fufh, cred, pid); 294 if (err) 295 return err; 296 297 fdisp_init(&fdi, sizeof(*ffi)); 298 fdisp_make_vp(&fdi, FUSE_FLUSH, vp, td, cred); 299 ffi = fdi.indata; 300 ffi->fh = fufh->fh_id; 301 /* 302 * If the file has a POSIX lock then we're supposed to set lock_owner. 303 * If not, then lock_owner is undefined. So we may as well always set 304 * it. 305 */ 306 ffi->lock_owner = td->td_proc->p_pid; 307 308 err = fdisp_wait_answ(&fdi); 309 if (err == ENOSYS) { 310 fsess_set_notimpl(mp, FUSE_FLUSH); 311 err = 0; 312 } 313 fdisp_destroy(&fdi); 314 return err; 315 } 316 317 /* Close wrapper for fifos. */ 318 static int 319 fuse_fifo_close(struct vop_close_args *ap) 320 { 321 return (fifo_specops.vop_close(ap)); 322 } 323 324 /* Invalidate a range of cached data, whether dirty of not */ 325 static int 326 fuse_inval_buf_range(struct vnode *vp, off_t filesize, off_t start, off_t end) 327 { 328 struct buf *bp; 329 daddr_t left_lbn, end_lbn, right_lbn; 330 off_t new_filesize; 331 int iosize, left_on, right_on, right_blksize; 332 333 iosize = fuse_iosize(vp); 334 left_lbn = start / iosize; 335 end_lbn = howmany(end, iosize); 336 left_on = start & (iosize - 1); 337 if (left_on != 0) { 338 bp = getblk(vp, left_lbn, iosize, PCATCH, 0, 0); 339 if ((bp->b_flags & B_CACHE) != 0 && bp->b_dirtyend >= left_on) { 340 /* 341 * Flush the dirty buffer, because we don't have a 342 * byte-granular way to record which parts of the 343 * buffer are valid. 344 */ 345 bwrite(bp); 346 if (bp->b_error) 347 return (bp->b_error); 348 } else { 349 brelse(bp); 350 } 351 } 352 right_on = end & (iosize - 1); 353 if (right_on != 0) { 354 right_lbn = end / iosize; 355 new_filesize = MAX(filesize, end); 356 right_blksize = MIN(iosize, new_filesize - iosize * right_lbn); 357 bp = getblk(vp, right_lbn, right_blksize, PCATCH, 0, 0); 358 if ((bp->b_flags & B_CACHE) != 0 && bp->b_dirtyoff < right_on) { 359 /* 360 * Flush the dirty buffer, because we don't have a 361 * byte-granular way to record which parts of the 362 * buffer are valid. 363 */ 364 bwrite(bp); 365 if (bp->b_error) 366 return (bp->b_error); 367 } else { 368 brelse(bp); 369 } 370 } 371 372 v_inval_buf_range(vp, left_lbn, end_lbn, iosize); 373 return (0); 374 } 375 376 377 /* Send FUSE_LSEEK for this node */ 378 static int 379 fuse_vnop_do_lseek(struct vnode *vp, struct thread *td, struct ucred *cred, 380 pid_t pid, off_t *offp, int whence) 381 { 382 struct fuse_dispatcher fdi; 383 struct fuse_filehandle *fufh; 384 struct fuse_lseek_in *flsi; 385 struct fuse_lseek_out *flso; 386 struct mount *mp = vnode_mount(vp); 387 int err; 388 389 ASSERT_VOP_LOCKED(vp, __func__); 390 391 err = fuse_filehandle_getrw(vp, FREAD, &fufh, cred, pid); 392 if (err) 393 return (err); 394 fdisp_init(&fdi, sizeof(*flsi)); 395 fdisp_make_vp(&fdi, FUSE_LSEEK, vp, td, cred); 396 flsi = fdi.indata; 397 flsi->fh = fufh->fh_id; 398 flsi->offset = *offp; 399 flsi->whence = whence; 400 err = fdisp_wait_answ(&fdi); 401 if (err == ENOSYS) { 402 fsess_set_notimpl(mp, FUSE_LSEEK); 403 } else if (err == 0) { 404 fsess_set_impl(mp, FUSE_LSEEK); 405 flso = fdi.answ; 406 *offp = flso->offset; 407 } 408 fdisp_destroy(&fdi); 409 410 return (err); 411 } 412 413 /* 414 struct vnop_access_args { 415 struct vnode *a_vp; 416 #if VOP_ACCESS_TAKES_ACCMODE_T 417 accmode_t a_accmode; 418 #else 419 int a_mode; 420 #endif 421 struct ucred *a_cred; 422 struct thread *a_td; 423 }; 424 */ 425 static int 426 fuse_vnop_access(struct vop_access_args *ap) 427 { 428 struct vnode *vp = ap->a_vp; 429 int accmode = ap->a_accmode; 430 struct ucred *cred = ap->a_cred; 431 432 struct fuse_data *data = fuse_get_mpdata(vnode_mount(vp)); 433 434 int err; 435 436 if (fuse_isdeadfs(vp)) { 437 if (vnode_isvroot(vp)) { 438 return 0; 439 } 440 return ENXIO; 441 } 442 if (!(data->dataflags & FSESS_INITED)) { 443 if (vnode_isvroot(vp)) { 444 if (priv_check_cred(cred, PRIV_VFS_ADMIN) || 445 (fuse_match_cred(data->daemoncred, cred) == 0)) { 446 return 0; 447 } 448 } 449 return EBADF; 450 } 451 if (vnode_islnk(vp)) { 452 return 0; 453 } 454 455 err = fuse_internal_access(vp, accmode, ap->a_td, ap->a_cred); 456 return err; 457 } 458 459 /* 460 * struct vop_advlock_args { 461 * struct vop_generic_args a_gen; 462 * struct vnode *a_vp; 463 * void *a_id; 464 * int a_op; 465 * struct flock *a_fl; 466 * int a_flags; 467 * } 468 */ 469 static int 470 fuse_vnop_advlock(struct vop_advlock_args *ap) 471 { 472 struct vnode *vp = ap->a_vp; 473 struct flock *fl = ap->a_fl; 474 struct thread *td = curthread; 475 struct ucred *cred = td->td_ucred; 476 pid_t pid = td->td_proc->p_pid; 477 struct fuse_filehandle *fufh; 478 struct fuse_dispatcher fdi; 479 struct fuse_lk_in *fli; 480 struct fuse_lk_out *flo; 481 enum fuse_opcode op; 482 int dataflags, err; 483 int flags = ap->a_flags; 484 485 dataflags = fuse_get_mpdata(vnode_mount(vp))->dataflags; 486 487 if (fuse_isdeadfs(vp)) { 488 return ENXIO; 489 } 490 491 switch(ap->a_op) { 492 case F_GETLK: 493 op = FUSE_GETLK; 494 break; 495 case F_SETLK: 496 if (flags & F_WAIT) 497 op = FUSE_SETLKW; 498 else 499 op = FUSE_SETLK; 500 break; 501 case F_UNLCK: 502 op = FUSE_SETLK; 503 break; 504 default: 505 return EINVAL; 506 } 507 508 if (!(dataflags & FSESS_POSIX_LOCKS)) 509 return vop_stdadvlock(ap); 510 /* FUSE doesn't properly support flock until protocol 7.17 */ 511 if (flags & F_FLOCK) 512 return vop_stdadvlock(ap); 513 514 vn_lock(vp, LK_SHARED | LK_RETRY); 515 516 err = fuse_filehandle_get_anyflags(vp, &fufh, cred, pid); 517 if (err) 518 goto out; 519 520 fdisp_init(&fdi, sizeof(*fli)); 521 522 fdisp_make_vp(&fdi, op, vp, td, cred); 523 fli = fdi.indata; 524 fli->fh = fufh->fh_id; 525 fli->owner = td->td_proc->p_pid; 526 fli->lk.start = fl->l_start; 527 if (fl->l_len != 0) 528 fli->lk.end = fl->l_start + fl->l_len - 1; 529 else 530 fli->lk.end = INT64_MAX; 531 fli->lk.type = fl->l_type; 532 fli->lk.pid = td->td_proc->p_pid; 533 534 err = fdisp_wait_answ(&fdi); 535 fdisp_destroy(&fdi); 536 537 if (err == 0 && op == FUSE_GETLK) { 538 flo = fdi.answ; 539 fl->l_type = flo->lk.type; 540 fl->l_pid = flo->lk.pid; 541 if (flo->lk.type != F_UNLCK) { 542 fl->l_start = flo->lk.start; 543 if (flo->lk.end == INT64_MAX) 544 fl->l_len = 0; 545 else 546 fl->l_len = flo->lk.end - flo->lk.start + 1; 547 fl->l_start = flo->lk.start; 548 } 549 } 550 551 out: 552 VOP_UNLOCK(vp); 553 return err; 554 } 555 556 static int 557 fuse_vnop_allocate(struct vop_allocate_args *ap) 558 { 559 struct vnode *vp = ap->a_vp; 560 off_t *len = ap->a_len; 561 off_t *offset = ap->a_offset; 562 struct ucred *cred = ap->a_cred; 563 struct fuse_filehandle *fufh; 564 struct mount *mp = vnode_mount(vp); 565 struct fuse_dispatcher fdi; 566 struct fuse_fallocate_in *ffi; 567 struct uio io; 568 pid_t pid = curthread->td_proc->p_pid; 569 struct fuse_vnode_data *fvdat = VTOFUD(vp); 570 off_t filesize; 571 int err; 572 573 if (fuse_isdeadfs(vp)) 574 return (ENXIO); 575 576 switch (vp->v_type) { 577 case VFIFO: 578 return (ESPIPE); 579 case VLNK: 580 case VREG: 581 if (vfs_isrdonly(mp)) 582 return (EROFS); 583 break; 584 default: 585 return (ENODEV); 586 } 587 588 if (vfs_isrdonly(mp)) 589 return (EROFS); 590 591 if (fsess_not_impl(mp, FUSE_FALLOCATE)) 592 return (EINVAL); 593 594 io.uio_offset = *offset; 595 io.uio_resid = *len; 596 err = vn_rlimit_fsize(vp, &io, curthread); 597 if (err) 598 return (err); 599 600 err = fuse_filehandle_getrw(vp, FWRITE, &fufh, cred, pid); 601 if (err) 602 return (err); 603 604 fuse_vnode_update(vp, FN_MTIMECHANGE | FN_CTIMECHANGE); 605 606 err = fuse_vnode_size(vp, &filesize, cred, curthread); 607 if (err) 608 return (err); 609 fuse_inval_buf_range(vp, filesize, *offset, *offset + *len); 610 611 fdisp_init(&fdi, sizeof(*ffi)); 612 fdisp_make_vp(&fdi, FUSE_FALLOCATE, vp, curthread, cred); 613 ffi = fdi.indata; 614 ffi->fh = fufh->fh_id; 615 ffi->offset = *offset; 616 ffi->length = *len; 617 ffi->mode = 0; 618 err = fdisp_wait_answ(&fdi); 619 620 if (err == ENOSYS) { 621 fsess_set_notimpl(mp, FUSE_FALLOCATE); 622 err = EINVAL; 623 } else if (err == EOPNOTSUPP) { 624 /* 625 * The file system server does not support FUSE_FALLOCATE with 626 * the supplied mode for this particular file. 627 */ 628 err = EINVAL; 629 } else if (!err) { 630 *offset += *len; 631 *len = 0; 632 fuse_vnode_undirty_cached_timestamps(vp, false); 633 fuse_internal_clear_suid_on_write(vp, cred, curthread); 634 if (*offset > fvdat->cached_attrs.va_size) { 635 fuse_vnode_setsize(vp, *offset, false); 636 getnanouptime(&fvdat->last_local_modify); 637 } 638 } 639 640 return (err); 641 } 642 643 /* { 644 struct vnode *a_vp; 645 daddr_t a_bn; 646 struct bufobj **a_bop; 647 daddr_t *a_bnp; 648 int *a_runp; 649 int *a_runb; 650 } */ 651 static int 652 fuse_vnop_bmap(struct vop_bmap_args *ap) 653 { 654 struct vnode *vp = ap->a_vp; 655 struct bufobj **bo = ap->a_bop; 656 struct thread *td = curthread; 657 struct mount *mp; 658 struct fuse_dispatcher fdi; 659 struct fuse_bmap_in *fbi; 660 struct fuse_bmap_out *fbo; 661 struct fuse_data *data; 662 struct fuse_vnode_data *fvdat = VTOFUD(vp); 663 uint64_t biosize; 664 off_t fsize; 665 daddr_t lbn = ap->a_bn; 666 daddr_t *pbn = ap->a_bnp; 667 int *runp = ap->a_runp; 668 int *runb = ap->a_runb; 669 int error = 0; 670 int maxrun; 671 672 if (fuse_isdeadfs(vp)) { 673 return ENXIO; 674 } 675 676 mp = vnode_mount(vp); 677 data = fuse_get_mpdata(mp); 678 biosize = fuse_iosize(vp); 679 maxrun = MIN(vp->v_mount->mnt_iosize_max / biosize - 1, 680 data->max_readahead_blocks); 681 682 if (bo != NULL) 683 *bo = &vp->v_bufobj; 684 685 /* 686 * The FUSE_BMAP operation does not include the runp and runb 687 * variables, so we must guess. Report nonzero contiguous runs so 688 * cluster_read will combine adjacent reads. It's worthwhile to reduce 689 * upcalls even if we don't know the true physical layout of the file. 690 * 691 * FUSE file systems may opt out of read clustering in two ways: 692 * * mounting with -onoclusterr 693 * * Setting max_readahead <= maxbcachebuf during FUSE_INIT 694 */ 695 if (runb != NULL) 696 *runb = MIN(lbn, maxrun); 697 if (runp != NULL && maxrun == 0) 698 *runp = 0; 699 else if (runp != NULL) { 700 /* 701 * If the file's size is cached, use that value to calculate 702 * runp, even if the cache is expired. runp is only advisory, 703 * and the risk of getting it wrong is not worth the cost of 704 * another upcall. 705 */ 706 if (fvdat->cached_attrs.va_size != VNOVAL) 707 fsize = fvdat->cached_attrs.va_size; 708 else 709 error = fuse_vnode_size(vp, &fsize, td->td_ucred, td); 710 if (error == 0) 711 *runp = MIN(MAX(0, fsize / (off_t)biosize - lbn - 1), 712 maxrun); 713 else 714 *runp = 0; 715 } 716 717 if (fsess_maybe_impl(mp, FUSE_BMAP)) { 718 fdisp_init(&fdi, sizeof(*fbi)); 719 fdisp_make_vp(&fdi, FUSE_BMAP, vp, td, td->td_ucred); 720 fbi = fdi.indata; 721 fbi->block = lbn; 722 fbi->blocksize = biosize; 723 error = fdisp_wait_answ(&fdi); 724 if (error == ENOSYS) { 725 fdisp_destroy(&fdi); 726 fsess_set_notimpl(mp, FUSE_BMAP); 727 error = 0; 728 } else { 729 fbo = fdi.answ; 730 if (error == 0 && pbn != NULL) 731 *pbn = fbo->block; 732 fdisp_destroy(&fdi); 733 return error; 734 } 735 } 736 737 /* If the daemon doesn't support BMAP, make up a sensible default */ 738 if (pbn != NULL) 739 *pbn = lbn * btodb(biosize); 740 return (error); 741 } 742 743 /* 744 struct vop_close_args { 745 struct vnode *a_vp; 746 int a_fflag; 747 struct ucred *a_cred; 748 struct thread *a_td; 749 }; 750 */ 751 static int 752 fuse_vnop_close(struct vop_close_args *ap) 753 { 754 struct vnode *vp = ap->a_vp; 755 struct ucred *cred = ap->a_cred; 756 int fflag = ap->a_fflag; 757 struct thread *td = ap->a_td; 758 pid_t pid = td->td_proc->p_pid; 759 struct fuse_vnode_data *fvdat = VTOFUD(vp); 760 int err = 0; 761 762 if (fuse_isdeadfs(vp)) 763 return 0; 764 if (vnode_isdir(vp)) 765 return 0; 766 if (fflag & IO_NDELAY) 767 return 0; 768 769 err = fuse_flush(vp, cred, pid, fflag); 770 if (err == 0 && (fvdat->flag & FN_ATIMECHANGE)) { 771 struct vattr vap; 772 773 VATTR_NULL(&vap); 774 vap.va_atime = fvdat->cached_attrs.va_atime; 775 err = fuse_internal_setattr(vp, &vap, td, NULL); 776 } 777 /* TODO: close the file handle, if we're sure it's no longer used */ 778 if ((fvdat->flag & FN_SIZECHANGE) != 0) { 779 fuse_vnode_savesize(vp, cred, td->td_proc->p_pid); 780 } 781 return err; 782 } 783 784 /* 785 struct vop_copy_file_range_args { 786 struct vop_generic_args a_gen; 787 struct vnode *a_invp; 788 off_t *a_inoffp; 789 struct vnode *a_outvp; 790 off_t *a_outoffp; 791 size_t *a_lenp; 792 unsigned int a_flags; 793 struct ucred *a_incred; 794 struct ucred *a_outcred; 795 struct thread *a_fsizetd; 796 } 797 */ 798 static int 799 fuse_vnop_copy_file_range(struct vop_copy_file_range_args *ap) 800 { 801 struct vnode *invp = ap->a_invp; 802 struct vnode *outvp = ap->a_outvp; 803 struct mount *mp = vnode_mount(invp); 804 struct fuse_vnode_data *outfvdat = VTOFUD(outvp); 805 struct fuse_dispatcher fdi; 806 struct fuse_filehandle *infufh, *outfufh; 807 struct fuse_copy_file_range_in *fcfri; 808 struct ucred *incred = ap->a_incred; 809 struct ucred *outcred = ap->a_outcred; 810 struct fuse_write_out *fwo; 811 struct thread *td; 812 struct uio io; 813 off_t outfilesize; 814 pid_t pid; 815 int err; 816 817 if (mp != vnode_mount(outvp)) 818 goto fallback; 819 820 if (incred->cr_uid != outcred->cr_uid) 821 goto fallback; 822 823 if (incred->cr_groups[0] != outcred->cr_groups[0]) 824 goto fallback; 825 826 if (fsess_not_impl(mp, FUSE_COPY_FILE_RANGE)) 827 goto fallback; 828 829 if (ap->a_fsizetd == NULL) 830 td = curthread; 831 else 832 td = ap->a_fsizetd; 833 pid = td->td_proc->p_pid; 834 835 /* Lock both vnodes, avoiding risk of deadlock. */ 836 do { 837 err = vn_lock(outvp, LK_EXCLUSIVE); 838 if (invp == outvp) 839 break; 840 if (err == 0) { 841 err = vn_lock(invp, LK_SHARED | LK_NOWAIT); 842 if (err == 0) 843 break; 844 VOP_UNLOCK(outvp); 845 err = vn_lock(invp, LK_SHARED); 846 if (err == 0) 847 VOP_UNLOCK(invp); 848 } 849 } while (err == 0); 850 if (err != 0) 851 return (err); 852 853 err = fuse_filehandle_getrw(invp, FREAD, &infufh, incred, pid); 854 if (err) 855 goto unlock; 856 857 err = fuse_filehandle_getrw(outvp, FWRITE, &outfufh, outcred, pid); 858 if (err) 859 goto unlock; 860 861 if (ap->a_fsizetd) { 862 io.uio_offset = *ap->a_outoffp; 863 io.uio_resid = *ap->a_lenp; 864 err = vn_rlimit_fsize(outvp, &io, ap->a_fsizetd); 865 if (err) 866 goto unlock; 867 } 868 869 err = fuse_vnode_size(outvp, &outfilesize, outcred, curthread); 870 if (err) 871 goto unlock; 872 873 err = fuse_inval_buf_range(outvp, outfilesize, *ap->a_outoffp, 874 *ap->a_outoffp + *ap->a_lenp); 875 if (err) 876 goto unlock; 877 878 fdisp_init(&fdi, sizeof(*fcfri)); 879 fdisp_make_vp(&fdi, FUSE_COPY_FILE_RANGE, invp, td, incred); 880 fcfri = fdi.indata; 881 fcfri->fh_in = infufh->fh_id; 882 fcfri->off_in = *ap->a_inoffp; 883 fcfri->nodeid_out = VTOI(outvp); 884 fcfri->fh_out = outfufh->fh_id; 885 fcfri->off_out = *ap->a_outoffp; 886 fcfri->len = *ap->a_lenp; 887 fcfri->flags = 0; 888 889 err = fdisp_wait_answ(&fdi); 890 if (err == 0) { 891 fwo = fdi.answ; 892 *ap->a_lenp = fwo->size; 893 *ap->a_inoffp += fwo->size; 894 *ap->a_outoffp += fwo->size; 895 fuse_internal_clear_suid_on_write(outvp, outcred, td); 896 if (*ap->a_outoffp > outfvdat->cached_attrs.va_size) { 897 fuse_vnode_setsize(outvp, *ap->a_outoffp, false); 898 getnanouptime(&outfvdat->last_local_modify); 899 } 900 fuse_vnode_update(invp, FN_ATIMECHANGE); 901 fuse_vnode_update(outvp, FN_MTIMECHANGE | FN_CTIMECHANGE); 902 } 903 fdisp_destroy(&fdi); 904 905 unlock: 906 if (invp != outvp) 907 VOP_UNLOCK(invp); 908 VOP_UNLOCK(outvp); 909 910 if (err == ENOSYS) { 911 fsess_set_notimpl(mp, FUSE_COPY_FILE_RANGE); 912 fallback: 913 err = vn_generic_copy_file_range(ap->a_invp, ap->a_inoffp, 914 ap->a_outvp, ap->a_outoffp, ap->a_lenp, ap->a_flags, 915 ap->a_incred, ap->a_outcred, ap->a_fsizetd); 916 } 917 918 return (err); 919 } 920 921 static void 922 fdisp_make_mknod_for_fallback( 923 struct fuse_dispatcher *fdip, 924 struct componentname *cnp, 925 struct vnode *dvp, 926 uint64_t parentnid, 927 struct thread *td, 928 struct ucred *cred, 929 mode_t mode, 930 enum fuse_opcode *op) 931 { 932 struct fuse_mknod_in *fmni; 933 934 fdisp_init(fdip, sizeof(*fmni) + cnp->cn_namelen + 1); 935 *op = FUSE_MKNOD; 936 fdisp_make(fdip, *op, vnode_mount(dvp), parentnid, td, cred); 937 fmni = fdip->indata; 938 fmni->mode = mode; 939 fmni->rdev = 0; 940 memcpy((char *)fdip->indata + sizeof(*fmni), cnp->cn_nameptr, 941 cnp->cn_namelen); 942 ((char *)fdip->indata)[sizeof(*fmni) + cnp->cn_namelen] = '\0'; 943 } 944 /* 945 struct vnop_create_args { 946 struct vnode *a_dvp; 947 struct vnode **a_vpp; 948 struct componentname *a_cnp; 949 struct vattr *a_vap; 950 }; 951 */ 952 static int 953 fuse_vnop_create(struct vop_create_args *ap) 954 { 955 struct vnode *dvp = ap->a_dvp; 956 struct vnode **vpp = ap->a_vpp; 957 struct componentname *cnp = ap->a_cnp; 958 struct vattr *vap = ap->a_vap; 959 struct thread *td = curthread; 960 struct ucred *cred = cnp->cn_cred; 961 962 struct fuse_data *data; 963 struct fuse_create_in *fci; 964 struct fuse_entry_out *feo; 965 struct fuse_open_out *foo; 966 struct fuse_dispatcher fdi, fdi2; 967 struct fuse_dispatcher *fdip = &fdi; 968 struct fuse_dispatcher *fdip2 = NULL; 969 970 int err; 971 972 struct mount *mp = vnode_mount(dvp); 973 data = fuse_get_mpdata(mp); 974 uint64_t parentnid = VTOFUD(dvp)->nid; 975 mode_t mode = MAKEIMODE(vap->va_type, vap->va_mode); 976 enum fuse_opcode op; 977 int flags; 978 979 if (fuse_isdeadfs(dvp)) 980 return ENXIO; 981 982 /* FUSE expects sockets to be created with FUSE_MKNOD */ 983 if (vap->va_type == VSOCK) 984 return fuse_internal_mknod(dvp, vpp, cnp, vap); 985 986 /* 987 * VOP_CREATE doesn't tell us the open(2) flags, so we guess. Only a 988 * writable mode makes sense, and we might as well include readability 989 * too. 990 */ 991 flags = O_RDWR; 992 993 bzero(&fdi, sizeof(fdi)); 994 995 if (vap->va_type != VREG) 996 return (EINVAL); 997 998 if (fsess_not_impl(mp, FUSE_CREATE) || vap->va_type == VSOCK) { 999 /* Fallback to FUSE_MKNOD/FUSE_OPEN */ 1000 fdisp_make_mknod_for_fallback(fdip, cnp, dvp, parentnid, td, 1001 cred, mode, &op); 1002 } else { 1003 /* Use FUSE_CREATE */ 1004 size_t insize; 1005 1006 op = FUSE_CREATE; 1007 fdisp_init(fdip, sizeof(*fci) + cnp->cn_namelen + 1); 1008 fdisp_make(fdip, op, vnode_mount(dvp), parentnid, td, cred); 1009 fci = fdip->indata; 1010 fci->mode = mode; 1011 fci->flags = O_CREAT | flags; 1012 if (fuse_libabi_geq(data, 7, 12)) { 1013 insize = sizeof(*fci); 1014 fci->umask = td->td_proc->p_pd->pd_cmask; 1015 } else { 1016 insize = sizeof(struct fuse_open_in); 1017 } 1018 1019 memcpy((char *)fdip->indata + insize, cnp->cn_nameptr, 1020 cnp->cn_namelen); 1021 ((char *)fdip->indata)[insize + cnp->cn_namelen] = '\0'; 1022 } 1023 1024 err = fdisp_wait_answ(fdip); 1025 1026 if (err) { 1027 if (err == ENOSYS && op == FUSE_CREATE) { 1028 fsess_set_notimpl(mp, FUSE_CREATE); 1029 fdisp_destroy(fdip); 1030 fdisp_make_mknod_for_fallback(fdip, cnp, dvp, 1031 parentnid, td, cred, mode, &op); 1032 err = fdisp_wait_answ(fdip); 1033 } 1034 if (err) 1035 goto out; 1036 } 1037 1038 feo = fdip->answ; 1039 1040 if ((err = fuse_internal_checkentry(feo, vap->va_type))) { 1041 goto out; 1042 } 1043 1044 if (op == FUSE_CREATE) { 1045 if (fuse_libabi_geq(data, 7, 9)) 1046 foo = (struct fuse_open_out*)(feo + 1); 1047 else 1048 foo = (struct fuse_open_out*)((char*)feo + 1049 FUSE_COMPAT_ENTRY_OUT_SIZE); 1050 } else { 1051 /* Issue a separate FUSE_OPEN */ 1052 struct fuse_open_in *foi; 1053 1054 fdip2 = &fdi2; 1055 fdisp_init(fdip2, sizeof(*foi)); 1056 fdisp_make(fdip2, FUSE_OPEN, vnode_mount(dvp), feo->nodeid, td, 1057 cred); 1058 foi = fdip2->indata; 1059 foi->flags = flags; 1060 err = fdisp_wait_answ(fdip2); 1061 if (err) 1062 goto out; 1063 foo = fdip2->answ; 1064 } 1065 err = fuse_vnode_get(mp, feo, feo->nodeid, dvp, vpp, cnp, vap->va_type); 1066 if (err) { 1067 struct fuse_release_in *fri; 1068 uint64_t nodeid = feo->nodeid; 1069 uint64_t fh_id = foo->fh; 1070 1071 fdisp_init(fdip, sizeof(*fri)); 1072 fdisp_make(fdip, FUSE_RELEASE, mp, nodeid, td, cred); 1073 fri = fdip->indata; 1074 fri->fh = fh_id; 1075 fri->flags = flags; 1076 fuse_insert_callback(fdip->tick, fuse_internal_forget_callback); 1077 fuse_insert_message(fdip->tick, false); 1078 goto out; 1079 } 1080 ASSERT_VOP_ELOCKED(*vpp, "fuse_vnop_create"); 1081 fuse_internal_cache_attrs(*vpp, &feo->attr, feo->attr_valid, 1082 feo->attr_valid_nsec, NULL, true); 1083 1084 fuse_filehandle_init(*vpp, FUFH_RDWR, NULL, td, cred, foo); 1085 fuse_vnode_open(*vpp, foo->open_flags, td); 1086 /* 1087 * Purge the parent's attribute cache because the daemon should've 1088 * updated its mtime and ctime 1089 */ 1090 fuse_vnode_clear_attr_cache(dvp); 1091 cache_purge_negative(dvp); 1092 1093 out: 1094 if (fdip2) 1095 fdisp_destroy(fdip2); 1096 fdisp_destroy(fdip); 1097 return err; 1098 } 1099 1100 /* 1101 struct vnop_fdatasync_args { 1102 struct vop_generic_args a_gen; 1103 struct vnode * a_vp; 1104 struct thread * a_td; 1105 }; 1106 */ 1107 static int 1108 fuse_vnop_fdatasync(struct vop_fdatasync_args *ap) 1109 { 1110 struct vnode *vp = ap->a_vp; 1111 struct thread *td = ap->a_td; 1112 int waitfor = MNT_WAIT; 1113 1114 int err = 0; 1115 1116 if (fuse_isdeadfs(vp)) { 1117 return 0; 1118 } 1119 if ((err = vop_stdfdatasync_buf(ap))) 1120 return err; 1121 1122 return fuse_internal_fsync(vp, td, waitfor, true); 1123 } 1124 1125 /* 1126 struct vnop_fsync_args { 1127 struct vop_generic_args a_gen; 1128 struct vnode * a_vp; 1129 int a_waitfor; 1130 struct thread * a_td; 1131 }; 1132 */ 1133 static int 1134 fuse_vnop_fsync(struct vop_fsync_args *ap) 1135 { 1136 struct vnode *vp = ap->a_vp; 1137 struct thread *td = ap->a_td; 1138 int waitfor = ap->a_waitfor; 1139 int err = 0; 1140 1141 if (fuse_isdeadfs(vp)) { 1142 return 0; 1143 } 1144 if ((err = vop_stdfsync(ap))) 1145 return err; 1146 1147 return fuse_internal_fsync(vp, td, waitfor, false); 1148 } 1149 1150 /* 1151 struct vnop_getattr_args { 1152 struct vnode *a_vp; 1153 struct vattr *a_vap; 1154 struct ucred *a_cred; 1155 struct thread *a_td; 1156 }; 1157 */ 1158 static int 1159 fuse_vnop_getattr(struct vop_getattr_args *ap) 1160 { 1161 struct vnode *vp = ap->a_vp; 1162 struct vattr *vap = ap->a_vap; 1163 struct ucred *cred = ap->a_cred; 1164 struct thread *td = curthread; 1165 1166 int err = 0; 1167 int dataflags; 1168 1169 dataflags = fuse_get_mpdata(vnode_mount(vp))->dataflags; 1170 1171 /* Note that we are not bailing out on a dead file system just yet. */ 1172 1173 if (!(dataflags & FSESS_INITED)) { 1174 if (!vnode_isvroot(vp)) { 1175 fdata_set_dead(fuse_get_mpdata(vnode_mount(vp))); 1176 err = ENOTCONN; 1177 return err; 1178 } else { 1179 goto fake; 1180 } 1181 } 1182 err = fuse_internal_getattr(vp, vap, cred, td); 1183 if (err == ENOTCONN && vnode_isvroot(vp)) { 1184 /* see comment in fuse_vfsop_statfs() */ 1185 goto fake; 1186 } else { 1187 return err; 1188 } 1189 1190 fake: 1191 bzero(vap, sizeof(*vap)); 1192 vap->va_type = vnode_vtype(vp); 1193 1194 return 0; 1195 } 1196 1197 /* 1198 struct vnop_inactive_args { 1199 struct vnode *a_vp; 1200 }; 1201 */ 1202 static int 1203 fuse_vnop_inactive(struct vop_inactive_args *ap) 1204 { 1205 struct vnode *vp = ap->a_vp; 1206 struct thread *td = curthread; 1207 1208 struct fuse_vnode_data *fvdat = VTOFUD(vp); 1209 struct fuse_filehandle *fufh, *fufh_tmp; 1210 1211 int need_flush = 1; 1212 1213 LIST_FOREACH_SAFE(fufh, &fvdat->handles, next, fufh_tmp) { 1214 if (need_flush && vp->v_type == VREG) { 1215 if ((VTOFUD(vp)->flag & FN_SIZECHANGE) != 0) { 1216 fuse_vnode_savesize(vp, NULL, 0); 1217 } 1218 if ((fvdat->flag & FN_REVOKED) != 0) 1219 fuse_io_invalbuf(vp, td); 1220 else 1221 fuse_io_flushbuf(vp, MNT_WAIT, td); 1222 need_flush = 0; 1223 } 1224 fuse_filehandle_close(vp, fufh, td, NULL); 1225 } 1226 1227 if ((fvdat->flag & FN_REVOKED) != 0) 1228 vrecycle(vp); 1229 1230 return 0; 1231 } 1232 1233 /* 1234 struct vnop_ioctl_args { 1235 struct vnode *a_vp; 1236 u_long a_command; 1237 caddr_t a_data; 1238 int a_fflag; 1239 struct ucred *a_cred; 1240 struct thread *a_td; 1241 }; 1242 */ 1243 static int 1244 fuse_vnop_ioctl(struct vop_ioctl_args *ap) 1245 { 1246 struct vnode *vp = ap->a_vp; 1247 struct mount *mp = vnode_mount(vp); 1248 struct ucred *cred = ap->a_cred; 1249 off_t *offp; 1250 pid_t pid = ap->a_td->td_proc->p_pid; 1251 int err; 1252 1253 switch (ap->a_command) { 1254 case FIOSEEKDATA: 1255 case FIOSEEKHOLE: 1256 /* Call FUSE_LSEEK, if we can, or fall back to vop_stdioctl */ 1257 if (fsess_maybe_impl(mp, FUSE_LSEEK)) { 1258 int whence; 1259 1260 offp = ap->a_data; 1261 if (ap->a_command == FIOSEEKDATA) 1262 whence = SEEK_DATA; 1263 else 1264 whence = SEEK_HOLE; 1265 1266 vn_lock(vp, LK_SHARED | LK_RETRY); 1267 err = fuse_vnop_do_lseek(vp, ap->a_td, cred, pid, offp, 1268 whence); 1269 VOP_UNLOCK(vp); 1270 } 1271 if (fsess_not_impl(mp, FUSE_LSEEK)) 1272 err = vop_stdioctl(ap); 1273 break; 1274 default: 1275 /* TODO: implement FUSE_IOCTL */ 1276 err = ENOTTY; 1277 break; 1278 } 1279 return (err); 1280 } 1281 1282 1283 /* 1284 struct vnop_link_args { 1285 struct vnode *a_tdvp; 1286 struct vnode *a_vp; 1287 struct componentname *a_cnp; 1288 }; 1289 */ 1290 static int 1291 fuse_vnop_link(struct vop_link_args *ap) 1292 { 1293 struct vnode *vp = ap->a_vp; 1294 struct vnode *tdvp = ap->a_tdvp; 1295 struct componentname *cnp = ap->a_cnp; 1296 1297 struct vattr *vap = VTOVA(vp); 1298 1299 struct fuse_dispatcher fdi; 1300 struct fuse_entry_out *feo; 1301 struct fuse_link_in fli; 1302 1303 int err; 1304 1305 if (fuse_isdeadfs(vp)) { 1306 return ENXIO; 1307 } 1308 if (vnode_mount(tdvp) != vnode_mount(vp)) { 1309 return EXDEV; 1310 } 1311 1312 /* 1313 * This is a seatbelt check to protect naive userspace filesystems from 1314 * themselves and the limitations of the FUSE IPC protocol. If a 1315 * filesystem does not allow attribute caching, assume it is capable of 1316 * validating that nlink does not overflow. 1317 */ 1318 if (vap != NULL && vap->va_nlink >= FUSE_LINK_MAX) 1319 return EMLINK; 1320 fli.oldnodeid = VTOI(vp); 1321 1322 fdisp_init(&fdi, 0); 1323 fuse_internal_newentry_makerequest(vnode_mount(tdvp), VTOI(tdvp), cnp, 1324 FUSE_LINK, &fli, sizeof(fli), &fdi); 1325 if ((err = fdisp_wait_answ(&fdi))) { 1326 goto out; 1327 } 1328 feo = fdi.answ; 1329 1330 if (fli.oldnodeid != feo->nodeid) { 1331 struct fuse_data *data = fuse_get_mpdata(vnode_mount(vp)); 1332 fuse_warn(data, FSESS_WARN_ILLEGAL_INODE, 1333 "Assigned wrong inode for a hard link."); 1334 fuse_vnode_clear_attr_cache(vp); 1335 fuse_vnode_clear_attr_cache(tdvp); 1336 err = EIO; 1337 goto out; 1338 } 1339 1340 err = fuse_internal_checkentry(feo, vnode_vtype(vp)); 1341 if (!err) { 1342 /* 1343 * Purge the parent's attribute cache because the daemon 1344 * should've updated its mtime and ctime 1345 */ 1346 fuse_vnode_clear_attr_cache(tdvp); 1347 fuse_internal_cache_attrs(vp, &feo->attr, feo->attr_valid, 1348 feo->attr_valid_nsec, NULL, true); 1349 } 1350 out: 1351 fdisp_destroy(&fdi); 1352 return err; 1353 } 1354 1355 struct fuse_lookup_alloc_arg { 1356 struct fuse_entry_out *feo; 1357 struct componentname *cnp; 1358 uint64_t nid; 1359 enum vtype vtyp; 1360 }; 1361 1362 /* Callback for vn_get_ino */ 1363 static int 1364 fuse_lookup_alloc(struct mount *mp, void *arg, int lkflags, struct vnode **vpp) 1365 { 1366 struct fuse_lookup_alloc_arg *flaa = arg; 1367 1368 return fuse_vnode_get(mp, flaa->feo, flaa->nid, NULL, vpp, flaa->cnp, 1369 flaa->vtyp); 1370 } 1371 1372 SDT_PROBE_DEFINE3(fusefs, , vnops, cache_lookup, 1373 "int", "struct timespec*", "struct timespec*"); 1374 /* 1375 struct vnop_lookup_args { 1376 struct vnodeop_desc *a_desc; 1377 struct vnode *a_dvp; 1378 struct vnode **a_vpp; 1379 struct componentname *a_cnp; 1380 }; 1381 */ 1382 int 1383 fuse_vnop_lookup(struct vop_lookup_args *ap) 1384 { 1385 struct vnode *dvp = ap->a_dvp; 1386 struct vnode **vpp = ap->a_vpp; 1387 struct componentname *cnp = ap->a_cnp; 1388 struct thread *td = curthread; 1389 struct ucred *cred = cnp->cn_cred; 1390 struct timespec now; 1391 1392 int nameiop = cnp->cn_nameiop; 1393 int flags = cnp->cn_flags; 1394 int islastcn = flags & ISLASTCN; 1395 struct mount *mp = vnode_mount(dvp); 1396 struct fuse_data *data = fuse_get_mpdata(mp); 1397 int default_permissions = data->dataflags & FSESS_DEFAULT_PERMISSIONS; 1398 bool is_dot; 1399 1400 int err = 0; 1401 int lookup_err = 0; 1402 struct vnode *vp = NULL; 1403 1404 struct fuse_dispatcher fdi; 1405 bool did_lookup = false; 1406 struct fuse_entry_out *feo = NULL; 1407 enum vtype vtyp; /* vnode type of target */ 1408 1409 uint64_t nid; 1410 1411 if (fuse_isdeadfs(dvp)) { 1412 *vpp = NULL; 1413 return ENXIO; 1414 } 1415 if (!vnode_isdir(dvp)) 1416 return ENOTDIR; 1417 1418 if (islastcn && vfs_isrdonly(mp) && (nameiop != LOOKUP)) 1419 return EROFS; 1420 1421 if ((cnp->cn_flags & NOEXECCHECK) != 0) 1422 cnp->cn_flags &= ~NOEXECCHECK; 1423 else if ((err = fuse_internal_access(dvp, VEXEC, td, cred))) 1424 return err; 1425 1426 is_dot = cnp->cn_namelen == 1 && *(cnp->cn_nameptr) == '.'; 1427 if ((flags & ISDOTDOT) && !(data->dataflags & FSESS_EXPORT_SUPPORT)) 1428 { 1429 if (!(VTOFUD(dvp)->flag & FN_PARENT_NID)) { 1430 /* 1431 * Since the file system doesn't support ".." lookups, 1432 * we have no way to find this entry. 1433 */ 1434 return ESTALE; 1435 } 1436 nid = VTOFUD(dvp)->parent_nid; 1437 if (nid == 0) 1438 return ENOENT; 1439 /* .. is obviously a directory */ 1440 vtyp = VDIR; 1441 } else if (is_dot) { 1442 nid = VTOI(dvp); 1443 /* . is obviously a directory */ 1444 vtyp = VDIR; 1445 } else { 1446 struct timespec timeout; 1447 int ncpticks; /* here to accommodate for API contract */ 1448 1449 err = cache_lookup(dvp, vpp, cnp, &timeout, &ncpticks); 1450 getnanouptime(&now); 1451 SDT_PROBE3(fusefs, , vnops, cache_lookup, err, &timeout, &now); 1452 switch (err) { 1453 case -1: /* positive match */ 1454 if (timespeccmp(&timeout, &now, >)) { 1455 counter_u64_add(fuse_lookup_cache_hits, 1); 1456 } else { 1457 /* Cache timeout */ 1458 counter_u64_add(fuse_lookup_cache_misses, 1); 1459 bintime_clear( 1460 &VTOFUD(*vpp)->entry_cache_timeout); 1461 cache_purge(*vpp); 1462 if (dvp != *vpp) 1463 vput(*vpp); 1464 else 1465 vrele(*vpp); 1466 *vpp = NULL; 1467 break; 1468 } 1469 return 0; 1470 1471 case 0: /* no match in cache */ 1472 counter_u64_add(fuse_lookup_cache_misses, 1); 1473 break; 1474 1475 case ENOENT: /* negative match */ 1476 if (timespeccmp(&timeout, &now, <=)) { 1477 /* Cache timeout */ 1478 cache_purge_negative(dvp); 1479 break; 1480 } 1481 /* fall through */ 1482 default: 1483 return err; 1484 } 1485 1486 fdisp_init(&fdi, cnp->cn_namelen + 1); 1487 fdisp_make(&fdi, FUSE_LOOKUP, mp, VTOI(dvp), td, cred); 1488 1489 memcpy(fdi.indata, cnp->cn_nameptr, cnp->cn_namelen); 1490 ((char *)fdi.indata)[cnp->cn_namelen] = '\0'; 1491 lookup_err = fdisp_wait_answ(&fdi); 1492 did_lookup = true; 1493 1494 if (!lookup_err) { 1495 /* lookup call succeeded */ 1496 feo = (struct fuse_entry_out *)fdi.answ; 1497 nid = feo->nodeid; 1498 if (nid == 0) { 1499 /* zero nodeid means ENOENT and cache it */ 1500 struct timespec timeout; 1501 1502 fdi.answ_stat = ENOENT; 1503 lookup_err = ENOENT; 1504 if (cnp->cn_flags & MAKEENTRY) { 1505 fuse_validity_2_timespec(feo, &timeout); 1506 /* Use the same entry_time for .. as for 1507 * the file itself. That doesn't honor 1508 * exactly what the fuse server tells 1509 * us, but to do otherwise would require 1510 * another cache lookup at this point. 1511 */ 1512 struct timespec *dtsp = NULL; 1513 cache_enter_time(dvp, *vpp, cnp, 1514 &timeout, dtsp); 1515 } 1516 } 1517 vtyp = IFTOVT(feo->attr.mode); 1518 } 1519 if (lookup_err && (!fdi.answ_stat || lookup_err != ENOENT)) { 1520 fdisp_destroy(&fdi); 1521 return lookup_err; 1522 } 1523 } 1524 /* lookup_err, if non-zero, must be ENOENT at this point */ 1525 1526 if (lookup_err) { 1527 /* Entry not found */ 1528 if ((nameiop == CREATE || nameiop == RENAME) && islastcn) { 1529 if (default_permissions) 1530 err = fuse_internal_access(dvp, VWRITE, td, 1531 cred); 1532 else 1533 err = 0; 1534 if (!err) { 1535 err = EJUSTRETURN; 1536 } 1537 } else { 1538 err = ENOENT; 1539 } 1540 } else { 1541 /* Entry was found */ 1542 if (flags & ISDOTDOT) { 1543 struct fuse_lookup_alloc_arg flaa; 1544 1545 flaa.nid = nid; 1546 flaa.feo = feo; 1547 flaa.cnp = cnp; 1548 flaa.vtyp = vtyp; 1549 err = vn_vget_ino_gen(dvp, fuse_lookup_alloc, &flaa, 0, 1550 &vp); 1551 *vpp = vp; 1552 } else if (nid == VTOI(dvp)) { 1553 if (is_dot) { 1554 vref(dvp); 1555 *vpp = dvp; 1556 } else { 1557 fuse_warn(fuse_get_mpdata(mp), 1558 FSESS_WARN_ILLEGAL_INODE, 1559 "Assigned same inode to both parent and " 1560 "child."); 1561 err = EIO; 1562 } 1563 1564 } else { 1565 struct fuse_vnode_data *fvdat; 1566 1567 err = fuse_vnode_get(vnode_mount(dvp), feo, nid, dvp, 1568 &vp, cnp, vtyp); 1569 if (err) 1570 goto out; 1571 *vpp = vp; 1572 fvdat = VTOFUD(vp); 1573 1574 MPASS(feo != NULL); 1575 if (timespeccmp(&now, &fvdat->last_local_modify, >)) { 1576 /* 1577 * Attributes from the server are definitely 1578 * newer than the last attributes we sent to 1579 * the server, so cache them. 1580 */ 1581 fuse_internal_cache_attrs(*vpp, &feo->attr, 1582 feo->attr_valid, feo->attr_valid_nsec, 1583 NULL, true); 1584 } 1585 fuse_validity_2_bintime(feo->entry_valid, 1586 feo->entry_valid_nsec, 1587 &fvdat->entry_cache_timeout); 1588 1589 if ((nameiop == DELETE || nameiop == RENAME) && 1590 islastcn && default_permissions) 1591 { 1592 struct vattr dvattr; 1593 1594 err = fuse_internal_access(dvp, VWRITE, td, 1595 cred); 1596 if (err != 0) 1597 goto out; 1598 /* 1599 * if the parent's sticky bit is set, check 1600 * whether we're allowed to remove the file. 1601 * Need to figure out the vnode locking to make 1602 * this work. 1603 */ 1604 fuse_internal_getattr(dvp, &dvattr, cred, td); 1605 if ((dvattr.va_mode & S_ISTXT) && 1606 fuse_internal_access(dvp, VADMIN, td, 1607 cred) && 1608 fuse_internal_access(*vpp, VADMIN, td, 1609 cred)) { 1610 err = EPERM; 1611 goto out; 1612 } 1613 } 1614 } 1615 } 1616 out: 1617 if (err) { 1618 if (vp != NULL && dvp != vp) 1619 vput(vp); 1620 else if (vp != NULL) 1621 vrele(vp); 1622 *vpp = NULL; 1623 } 1624 if (did_lookup) 1625 fdisp_destroy(&fdi); 1626 1627 return err; 1628 } 1629 1630 /* 1631 struct vnop_mkdir_args { 1632 struct vnode *a_dvp; 1633 struct vnode **a_vpp; 1634 struct componentname *a_cnp; 1635 struct vattr *a_vap; 1636 }; 1637 */ 1638 static int 1639 fuse_vnop_mkdir(struct vop_mkdir_args *ap) 1640 { 1641 struct vnode *dvp = ap->a_dvp; 1642 struct vnode **vpp = ap->a_vpp; 1643 struct componentname *cnp = ap->a_cnp; 1644 struct vattr *vap = ap->a_vap; 1645 1646 struct fuse_mkdir_in fmdi; 1647 1648 if (fuse_isdeadfs(dvp)) { 1649 return ENXIO; 1650 } 1651 fmdi.mode = MAKEIMODE(vap->va_type, vap->va_mode); 1652 fmdi.umask = curthread->td_proc->p_pd->pd_cmask; 1653 1654 return (fuse_internal_newentry(dvp, vpp, cnp, FUSE_MKDIR, &fmdi, 1655 sizeof(fmdi), VDIR)); 1656 } 1657 1658 /* 1659 struct vnop_mknod_args { 1660 struct vnode *a_dvp; 1661 struct vnode **a_vpp; 1662 struct componentname *a_cnp; 1663 struct vattr *a_vap; 1664 }; 1665 */ 1666 static int 1667 fuse_vnop_mknod(struct vop_mknod_args *ap) 1668 { 1669 1670 struct vnode *dvp = ap->a_dvp; 1671 struct vnode **vpp = ap->a_vpp; 1672 struct componentname *cnp = ap->a_cnp; 1673 struct vattr *vap = ap->a_vap; 1674 1675 if (fuse_isdeadfs(dvp)) 1676 return ENXIO; 1677 1678 return fuse_internal_mknod(dvp, vpp, cnp, vap); 1679 } 1680 1681 /* 1682 struct vop_open_args { 1683 struct vnode *a_vp; 1684 int a_mode; 1685 struct ucred *a_cred; 1686 struct thread *a_td; 1687 int a_fdidx; / struct file *a_fp; 1688 }; 1689 */ 1690 static int 1691 fuse_vnop_open(struct vop_open_args *ap) 1692 { 1693 struct vnode *vp = ap->a_vp; 1694 int a_mode = ap->a_mode; 1695 struct thread *td = ap->a_td; 1696 struct ucred *cred = ap->a_cred; 1697 pid_t pid = td->td_proc->p_pid; 1698 1699 if (fuse_isdeadfs(vp)) 1700 return ENXIO; 1701 if (vp->v_type == VCHR || vp->v_type == VBLK || vp->v_type == VFIFO) 1702 return (EOPNOTSUPP); 1703 if ((a_mode & (FREAD | FWRITE | FEXEC)) == 0) 1704 return EINVAL; 1705 1706 if (fuse_filehandle_validrw(vp, a_mode, cred, pid)) { 1707 fuse_vnode_open(vp, 0, td); 1708 return 0; 1709 } 1710 1711 return fuse_filehandle_open(vp, a_mode, NULL, td, cred); 1712 } 1713 1714 static int 1715 fuse_vnop_pathconf(struct vop_pathconf_args *ap) 1716 { 1717 struct vnode *vp = ap->a_vp; 1718 struct mount *mp; 1719 1720 switch (ap->a_name) { 1721 case _PC_FILESIZEBITS: 1722 *ap->a_retval = 64; 1723 return (0); 1724 case _PC_NAME_MAX: 1725 *ap->a_retval = NAME_MAX; 1726 return (0); 1727 case _PC_LINK_MAX: 1728 *ap->a_retval = MIN(LONG_MAX, FUSE_LINK_MAX); 1729 return (0); 1730 case _PC_SYMLINK_MAX: 1731 *ap->a_retval = MAXPATHLEN; 1732 return (0); 1733 case _PC_NO_TRUNC: 1734 *ap->a_retval = 1; 1735 return (0); 1736 case _PC_MIN_HOLE_SIZE: 1737 /* 1738 * The FUSE protocol provides no mechanism for a server to 1739 * report _PC_MIN_HOLE_SIZE. It's a protocol bug. Instead, 1740 * return EINVAL if the server does not support FUSE_LSEEK, or 1741 * 1 if it does. 1742 */ 1743 mp = vnode_mount(vp); 1744 if (!fsess_is_impl(mp, FUSE_LSEEK) && 1745 !fsess_not_impl(mp, FUSE_LSEEK)) { 1746 off_t offset = 0; 1747 1748 /* Issue a FUSE_LSEEK to find out if it's implemented */ 1749 fuse_vnop_do_lseek(vp, curthread, curthread->td_ucred, 1750 curthread->td_proc->p_pid, &offset, SEEK_DATA); 1751 } 1752 1753 if (fsess_is_impl(mp, FUSE_LSEEK)) { 1754 *ap->a_retval = 1; 1755 return (0); 1756 } else { 1757 /* 1758 * Probably FUSE_LSEEK is not implemented. It might 1759 * be, if the FUSE_LSEEK above returned an error like 1760 * EACCES, but in that case we can't tell, so it's 1761 * safest to report EINVAL anyway. 1762 */ 1763 return (EINVAL); 1764 } 1765 default: 1766 return (vop_stdpathconf(ap)); 1767 } 1768 } 1769 1770 SDT_PROBE_DEFINE3(fusefs, , vnops, filehandles_closed, "struct vnode*", 1771 "struct uio*", "struct ucred*"); 1772 /* 1773 struct vnop_read_args { 1774 struct vnode *a_vp; 1775 struct uio *a_uio; 1776 int a_ioflag; 1777 struct ucred *a_cred; 1778 }; 1779 */ 1780 static int 1781 fuse_vnop_read(struct vop_read_args *ap) 1782 { 1783 struct vnode *vp = ap->a_vp; 1784 struct uio *uio = ap->a_uio; 1785 int ioflag = ap->a_ioflag; 1786 struct ucred *cred = ap->a_cred; 1787 pid_t pid = curthread->td_proc->p_pid; 1788 struct fuse_filehandle *fufh; 1789 int err; 1790 bool closefufh = false, directio; 1791 1792 MPASS(vp->v_type == VREG || vp->v_type == VDIR); 1793 1794 if (fuse_isdeadfs(vp)) { 1795 return ENXIO; 1796 } 1797 1798 if (VTOFUD(vp)->flag & FN_DIRECTIO) { 1799 ioflag |= IO_DIRECT; 1800 } 1801 1802 err = fuse_filehandle_getrw(vp, FREAD, &fufh, cred, pid); 1803 if (err == EBADF && vnode_mount(vp)->mnt_flag & MNT_EXPORTED) { 1804 /* 1805 * nfsd will do I/O without first doing VOP_OPEN. We 1806 * must implicitly open the file here 1807 */ 1808 err = fuse_filehandle_open(vp, FREAD, &fufh, curthread, cred); 1809 closefufh = true; 1810 } 1811 if (err) { 1812 SDT_PROBE3(fusefs, , vnops, filehandles_closed, vp, uio, cred); 1813 return err; 1814 } 1815 1816 /* 1817 * Ideally, when the daemon asks for direct io at open time, the 1818 * standard file flag should be set according to this, so that would 1819 * just change the default mode, which later on could be changed via 1820 * fcntl(2). 1821 * But this doesn't work, the O_DIRECT flag gets cleared at some point 1822 * (don't know where). So to make any use of the Fuse direct_io option, 1823 * we hardwire it into the file's private data (similarly to Linux, 1824 * btw.). 1825 */ 1826 directio = (ioflag & IO_DIRECT) || !fsess_opt_datacache(vnode_mount(vp)); 1827 1828 fuse_vnode_update(vp, FN_ATIMECHANGE); 1829 if (directio) { 1830 SDT_PROBE2(fusefs, , vnops, trace, 1, "direct read of vnode"); 1831 err = fuse_read_directbackend(vp, uio, cred, fufh); 1832 } else { 1833 SDT_PROBE2(fusefs, , vnops, trace, 1, "buffered read of vnode"); 1834 err = fuse_read_biobackend(vp, uio, ioflag, cred, fufh, pid); 1835 } 1836 1837 if (closefufh) 1838 fuse_filehandle_close(vp, fufh, curthread, cred); 1839 1840 return (err); 1841 } 1842 1843 /* 1844 struct vnop_readdir_args { 1845 struct vnode *a_vp; 1846 struct uio *a_uio; 1847 struct ucred *a_cred; 1848 int *a_eofflag; 1849 int *a_ncookies; 1850 uint64_t **a_cookies; 1851 }; 1852 */ 1853 static int 1854 fuse_vnop_readdir(struct vop_readdir_args *ap) 1855 { 1856 struct vnode *vp = ap->a_vp; 1857 struct uio *uio = ap->a_uio; 1858 struct ucred *cred = ap->a_cred; 1859 struct fuse_filehandle *fufh = NULL; 1860 struct mount *mp = vnode_mount(vp); 1861 struct fuse_iov cookediov; 1862 int err = 0; 1863 uint64_t *cookies; 1864 ssize_t tresid; 1865 int ncookies; 1866 bool closefufh = false; 1867 pid_t pid = curthread->td_proc->p_pid; 1868 1869 if (ap->a_eofflag) 1870 *ap->a_eofflag = 0; 1871 if (fuse_isdeadfs(vp)) { 1872 return ENXIO; 1873 } 1874 if ( /* XXXIP ((uio_iovcnt(uio) > 1)) || */ 1875 (uio_resid(uio) < sizeof(struct dirent))) { 1876 return EINVAL; 1877 } 1878 1879 tresid = uio->uio_resid; 1880 err = fuse_filehandle_get_dir(vp, &fufh, cred, pid); 1881 if (err == EBADF && mp->mnt_flag & MNT_EXPORTED) { 1882 KASSERT(fuse_get_mpdata(mp)->dataflags 1883 & FSESS_NO_OPENDIR_SUPPORT, 1884 ("FUSE file systems that don't set " 1885 "FUSE_NO_OPENDIR_SUPPORT should not be exported")); 1886 /* 1887 * nfsd will do VOP_READDIR without first doing VOP_OPEN. We 1888 * must implicitly open the directory here. 1889 */ 1890 err = fuse_filehandle_open(vp, FREAD, &fufh, curthread, cred); 1891 closefufh = true; 1892 } 1893 if (err) 1894 return (err); 1895 if (ap->a_ncookies != NULL) { 1896 ncookies = uio->uio_resid / 1897 (offsetof(struct dirent, d_name) + 4) + 1; 1898 cookies = malloc(ncookies * sizeof(*cookies), M_TEMP, M_WAITOK); 1899 *ap->a_ncookies = ncookies; 1900 *ap->a_cookies = cookies; 1901 } else { 1902 ncookies = 0; 1903 cookies = NULL; 1904 } 1905 #define DIRCOOKEDSIZE FUSE_DIRENT_ALIGN(FUSE_NAME_OFFSET + MAXNAMLEN + 1) 1906 fiov_init(&cookediov, DIRCOOKEDSIZE); 1907 1908 err = fuse_internal_readdir(vp, uio, fufh, &cookediov, 1909 &ncookies, cookies); 1910 1911 fiov_teardown(&cookediov); 1912 if (closefufh) 1913 fuse_filehandle_close(vp, fufh, curthread, cred); 1914 1915 if (ap->a_ncookies != NULL) { 1916 if (err == 0) { 1917 *ap->a_ncookies -= ncookies; 1918 } else { 1919 free(*ap->a_cookies, M_TEMP); 1920 *ap->a_ncookies = 0; 1921 *ap->a_cookies = NULL; 1922 } 1923 } 1924 if (err == 0 && tresid == uio->uio_resid) 1925 *ap->a_eofflag = 1; 1926 1927 return err; 1928 } 1929 1930 /* 1931 struct vnop_readlink_args { 1932 struct vnode *a_vp; 1933 struct uio *a_uio; 1934 struct ucred *a_cred; 1935 }; 1936 */ 1937 static int 1938 fuse_vnop_readlink(struct vop_readlink_args *ap) 1939 { 1940 struct vnode *vp = ap->a_vp; 1941 struct uio *uio = ap->a_uio; 1942 struct ucred *cred = ap->a_cred; 1943 1944 struct fuse_dispatcher fdi; 1945 int err; 1946 1947 if (fuse_isdeadfs(vp)) { 1948 return ENXIO; 1949 } 1950 if (!vnode_islnk(vp)) { 1951 return EINVAL; 1952 } 1953 fdisp_init(&fdi, 0); 1954 err = fdisp_simple_putget_vp(&fdi, FUSE_READLINK, vp, curthread, cred); 1955 if (err) { 1956 goto out; 1957 } 1958 if (((char *)fdi.answ)[0] == '/' && 1959 fuse_get_mpdata(vnode_mount(vp))->dataflags & FSESS_PUSH_SYMLINKS_IN) { 1960 char *mpth = vnode_mount(vp)->mnt_stat.f_mntonname; 1961 1962 err = uiomove(mpth, strlen(mpth), uio); 1963 } 1964 if (!err) { 1965 err = uiomove(fdi.answ, fdi.iosize, uio); 1966 } 1967 out: 1968 fdisp_destroy(&fdi); 1969 return err; 1970 } 1971 1972 /* 1973 struct vnop_reclaim_args { 1974 struct vnode *a_vp; 1975 }; 1976 */ 1977 static int 1978 fuse_vnop_reclaim(struct vop_reclaim_args *ap) 1979 { 1980 struct vnode *vp = ap->a_vp; 1981 struct thread *td = curthread; 1982 struct fuse_vnode_data *fvdat = VTOFUD(vp); 1983 struct fuse_filehandle *fufh, *fufh_tmp; 1984 1985 if (!fvdat) { 1986 panic("FUSE: no vnode data during recycling"); 1987 } 1988 LIST_FOREACH_SAFE(fufh, &fvdat->handles, next, fufh_tmp) { 1989 printf("FUSE: vnode being reclaimed with open fufh " 1990 "(type=%#x)", fufh->fufh_type); 1991 fuse_filehandle_close(vp, fufh, td, NULL); 1992 } 1993 1994 if (VTOI(vp) == 1) { 1995 /* 1996 * Don't send FUSE_FORGET for the root inode, because 1997 * we never send FUSE_LOOKUP for it (see 1998 * fuse_vfsop_root) and we don't want the server to see 1999 * mismatched lookup counts. 2000 */ 2001 struct fuse_data *data; 2002 struct vnode *vroot; 2003 2004 data = fuse_get_mpdata(vnode_mount(vp)); 2005 FUSE_LOCK(); 2006 vroot = data->vroot; 2007 data->vroot = NULL; 2008 FUSE_UNLOCK(); 2009 if (vroot) 2010 vrele(vroot); 2011 } else if (!fuse_isdeadfs(vp) && fvdat->nlookup > 0) { 2012 fuse_internal_forget_send(vnode_mount(vp), td, NULL, VTOI(vp), 2013 fvdat->nlookup); 2014 } 2015 cache_purge(vp); 2016 vfs_hash_remove(vp); 2017 fuse_vnode_destroy(vp); 2018 2019 return 0; 2020 } 2021 2022 /* 2023 struct vnop_remove_args { 2024 struct vnode *a_dvp; 2025 struct vnode *a_vp; 2026 struct componentname *a_cnp; 2027 }; 2028 */ 2029 static int 2030 fuse_vnop_remove(struct vop_remove_args *ap) 2031 { 2032 struct vnode *dvp = ap->a_dvp; 2033 struct vnode *vp = ap->a_vp; 2034 struct componentname *cnp = ap->a_cnp; 2035 2036 int err; 2037 2038 if (fuse_isdeadfs(vp)) { 2039 return ENXIO; 2040 } 2041 if (vnode_isdir(vp)) { 2042 return EPERM; 2043 } 2044 2045 err = fuse_internal_remove(dvp, vp, cnp, FUSE_UNLINK); 2046 2047 return err; 2048 } 2049 2050 /* 2051 struct vnop_rename_args { 2052 struct vnode *a_fdvp; 2053 struct vnode *a_fvp; 2054 struct componentname *a_fcnp; 2055 struct vnode *a_tdvp; 2056 struct vnode *a_tvp; 2057 struct componentname *a_tcnp; 2058 }; 2059 */ 2060 static int 2061 fuse_vnop_rename(struct vop_rename_args *ap) 2062 { 2063 struct vnode *fdvp = ap->a_fdvp; 2064 struct vnode *fvp = ap->a_fvp; 2065 struct componentname *fcnp = ap->a_fcnp; 2066 struct vnode *tdvp = ap->a_tdvp; 2067 struct vnode *tvp = ap->a_tvp; 2068 struct componentname *tcnp = ap->a_tcnp; 2069 struct fuse_data *data; 2070 bool newparent = fdvp != tdvp; 2071 bool isdir = fvp->v_type == VDIR; 2072 int err = 0; 2073 2074 if (fuse_isdeadfs(fdvp)) { 2075 return ENXIO; 2076 } 2077 if (fvp->v_mount != tdvp->v_mount || 2078 (tvp && fvp->v_mount != tvp->v_mount)) { 2079 SDT_PROBE2(fusefs, , vnops, trace, 1, "cross-device rename"); 2080 err = EXDEV; 2081 goto out; 2082 } 2083 cache_purge(fvp); 2084 2085 /* 2086 * FUSE library is expected to check if target directory is not 2087 * under the source directory in the file system tree. 2088 * Linux performs this check at VFS level. 2089 */ 2090 /* 2091 * If source is a directory, and it will get a new parent, user must 2092 * have write permission to it, so ".." can be modified. 2093 */ 2094 data = fuse_get_mpdata(vnode_mount(tdvp)); 2095 if (data->dataflags & FSESS_DEFAULT_PERMISSIONS && isdir && newparent) { 2096 err = fuse_internal_access(fvp, VWRITE, 2097 curthread, tcnp->cn_cred); 2098 if (err) 2099 goto out; 2100 } 2101 sx_xlock(&data->rename_lock); 2102 err = fuse_internal_rename(fdvp, fcnp, tdvp, tcnp); 2103 if (err == 0) { 2104 if (tdvp != fdvp) 2105 fuse_vnode_setparent(fvp, tdvp); 2106 if (tvp != NULL) 2107 fuse_vnode_setparent(tvp, NULL); 2108 } 2109 sx_unlock(&data->rename_lock); 2110 2111 if (tvp != NULL && tvp != fvp) { 2112 cache_purge(tvp); 2113 } 2114 if (vnode_isdir(fvp)) { 2115 if (((tvp != NULL) && vnode_isdir(tvp)) || vnode_isdir(fvp)) { 2116 cache_purge(tdvp); 2117 } 2118 cache_purge(fdvp); 2119 } 2120 out: 2121 if (tdvp == tvp) { 2122 vrele(tdvp); 2123 } else { 2124 vput(tdvp); 2125 } 2126 if (tvp != NULL) { 2127 vput(tvp); 2128 } 2129 vrele(fdvp); 2130 vrele(fvp); 2131 2132 return err; 2133 } 2134 2135 /* 2136 struct vnop_rmdir_args { 2137 struct vnode *a_dvp; 2138 struct vnode *a_vp; 2139 struct componentname *a_cnp; 2140 } *ap; 2141 */ 2142 static int 2143 fuse_vnop_rmdir(struct vop_rmdir_args *ap) 2144 { 2145 struct vnode *dvp = ap->a_dvp; 2146 struct vnode *vp = ap->a_vp; 2147 2148 int err; 2149 2150 if (fuse_isdeadfs(vp)) { 2151 return ENXIO; 2152 } 2153 if (VTOFUD(vp) == VTOFUD(dvp)) { 2154 return EINVAL; 2155 } 2156 err = fuse_internal_remove(dvp, vp, ap->a_cnp, FUSE_RMDIR); 2157 2158 return err; 2159 } 2160 2161 /* 2162 struct vnop_setattr_args { 2163 struct vnode *a_vp; 2164 struct vattr *a_vap; 2165 struct ucred *a_cred; 2166 struct thread *a_td; 2167 }; 2168 */ 2169 static int 2170 fuse_vnop_setattr(struct vop_setattr_args *ap) 2171 { 2172 struct vnode *vp = ap->a_vp; 2173 struct vattr *vap = ap->a_vap; 2174 struct ucred *cred = ap->a_cred; 2175 struct thread *td = curthread; 2176 struct mount *mp; 2177 struct fuse_data *data; 2178 struct vattr old_va; 2179 int dataflags; 2180 int err = 0, err2; 2181 accmode_t accmode = 0; 2182 bool checkperm; 2183 bool drop_suid = false; 2184 gid_t cr_gid; 2185 2186 mp = vnode_mount(vp); 2187 data = fuse_get_mpdata(mp); 2188 dataflags = data->dataflags; 2189 checkperm = dataflags & FSESS_DEFAULT_PERMISSIONS; 2190 if (cred->cr_ngroups > 0) 2191 cr_gid = cred->cr_groups[0]; 2192 else 2193 cr_gid = 0; 2194 2195 if (fuse_isdeadfs(vp)) { 2196 return ENXIO; 2197 } 2198 2199 if (vap->va_uid != (uid_t)VNOVAL) { 2200 if (checkperm) { 2201 /* Only root may change a file's owner */ 2202 err = priv_check_cred(cred, PRIV_VFS_CHOWN); 2203 if (err) { 2204 /* As a special case, allow the null chown */ 2205 err2 = fuse_internal_getattr(vp, &old_va, cred, 2206 td); 2207 if (err2) 2208 return (err2); 2209 if (vap->va_uid != old_va.va_uid) 2210 return err; 2211 else 2212 accmode |= VADMIN; 2213 drop_suid = true; 2214 } else 2215 accmode |= VADMIN; 2216 } else 2217 accmode |= VADMIN; 2218 } 2219 if (vap->va_gid != (gid_t)VNOVAL) { 2220 if (checkperm && priv_check_cred(cred, PRIV_VFS_CHOWN)) 2221 drop_suid = true; 2222 if (checkperm && !groupmember(vap->va_gid, cred)) 2223 { 2224 /* 2225 * Non-root users may only chgrp to one of their own 2226 * groups 2227 */ 2228 err = priv_check_cred(cred, PRIV_VFS_CHOWN); 2229 if (err) { 2230 /* As a special case, allow the null chgrp */ 2231 err2 = fuse_internal_getattr(vp, &old_va, cred, 2232 td); 2233 if (err2) 2234 return (err2); 2235 if (vap->va_gid != old_va.va_gid) 2236 return err; 2237 accmode |= VADMIN; 2238 } else 2239 accmode |= VADMIN; 2240 } else 2241 accmode |= VADMIN; 2242 } 2243 if (vap->va_size != VNOVAL) { 2244 switch (vp->v_type) { 2245 case VDIR: 2246 return (EISDIR); 2247 case VLNK: 2248 case VREG: 2249 if (vfs_isrdonly(mp)) 2250 return (EROFS); 2251 break; 2252 default: 2253 /* 2254 * According to POSIX, the result is unspecified 2255 * for file types other than regular files, 2256 * directories and shared memory objects. We 2257 * don't support shared memory objects in the file 2258 * system, and have dubious support for truncating 2259 * symlinks. Just ignore the request in other cases. 2260 */ 2261 return (0); 2262 } 2263 /* Don't set accmode. Permission to trunc is checked upstack */ 2264 } 2265 if (vap->va_atime.tv_sec != VNOVAL || vap->va_mtime.tv_sec != VNOVAL) { 2266 if (vap->va_vaflags & VA_UTIMES_NULL) 2267 accmode |= VWRITE; 2268 else 2269 accmode |= VADMIN; 2270 } 2271 if (drop_suid) { 2272 if (vap->va_mode != (mode_t)VNOVAL) 2273 vap->va_mode &= ~(S_ISUID | S_ISGID); 2274 else { 2275 err = fuse_internal_getattr(vp, &old_va, cred, td); 2276 if (err) 2277 return (err); 2278 vap->va_mode = old_va.va_mode & ~(S_ISUID | S_ISGID); 2279 } 2280 } 2281 if (vap->va_mode != (mode_t)VNOVAL) { 2282 /* Only root may set the sticky bit on non-directories */ 2283 if (checkperm && vp->v_type != VDIR && (vap->va_mode & S_ISTXT) 2284 && priv_check_cred(cred, PRIV_VFS_STICKYFILE)) 2285 return EFTYPE; 2286 if (checkperm && (vap->va_mode & S_ISGID)) { 2287 err = fuse_internal_getattr(vp, &old_va, cred, td); 2288 if (err) 2289 return (err); 2290 if (!groupmember(old_va.va_gid, cred)) { 2291 err = priv_check_cred(cred, PRIV_VFS_SETGID); 2292 if (err) 2293 return (err); 2294 } 2295 } 2296 accmode |= VADMIN; 2297 } 2298 2299 if (vfs_isrdonly(mp)) 2300 return EROFS; 2301 2302 if (checkperm) { 2303 err = fuse_internal_access(vp, accmode, td, cred); 2304 } else { 2305 err = 0; 2306 } 2307 if (err) 2308 return err; 2309 else 2310 return fuse_internal_setattr(vp, vap, td, cred); 2311 } 2312 2313 /* 2314 struct vnop_strategy_args { 2315 struct vnode *a_vp; 2316 struct buf *a_bp; 2317 }; 2318 */ 2319 static int 2320 fuse_vnop_strategy(struct vop_strategy_args *ap) 2321 { 2322 struct vnode *vp = ap->a_vp; 2323 struct buf *bp = ap->a_bp; 2324 2325 if (!vp || fuse_isdeadfs(vp)) { 2326 bp->b_ioflags |= BIO_ERROR; 2327 bp->b_error = ENXIO; 2328 bufdone(bp); 2329 return 0; 2330 } 2331 2332 /* 2333 * VOP_STRATEGY always returns zero and signals error via bp->b_ioflags. 2334 * fuse_io_strategy sets bp's error fields 2335 */ 2336 (void)fuse_io_strategy(vp, bp); 2337 2338 return 0; 2339 } 2340 2341 /* 2342 struct vnop_symlink_args { 2343 struct vnode *a_dvp; 2344 struct vnode **a_vpp; 2345 struct componentname *a_cnp; 2346 struct vattr *a_vap; 2347 char *a_target; 2348 }; 2349 */ 2350 static int 2351 fuse_vnop_symlink(struct vop_symlink_args *ap) 2352 { 2353 struct vnode *dvp = ap->a_dvp; 2354 struct vnode **vpp = ap->a_vpp; 2355 struct componentname *cnp = ap->a_cnp; 2356 const char *target = ap->a_target; 2357 2358 struct fuse_dispatcher fdi; 2359 2360 int err; 2361 size_t len; 2362 2363 if (fuse_isdeadfs(dvp)) { 2364 return ENXIO; 2365 } 2366 /* 2367 * Unlike the other creator type calls, here we have to create a message 2368 * where the name of the new entry comes first, and the data describing 2369 * the entry comes second. 2370 * Hence we can't rely on our handy fuse_internal_newentry() routine, 2371 * but put together the message manually and just call the core part. 2372 */ 2373 2374 len = strlen(target) + 1; 2375 fdisp_init(&fdi, len + cnp->cn_namelen + 1); 2376 fdisp_make_vp(&fdi, FUSE_SYMLINK, dvp, curthread, NULL); 2377 2378 memcpy(fdi.indata, cnp->cn_nameptr, cnp->cn_namelen); 2379 ((char *)fdi.indata)[cnp->cn_namelen] = '\0'; 2380 memcpy((char *)fdi.indata + cnp->cn_namelen + 1, target, len); 2381 2382 err = fuse_internal_newentry_core(dvp, vpp, cnp, VLNK, &fdi); 2383 fdisp_destroy(&fdi); 2384 return err; 2385 } 2386 2387 /* 2388 struct vnop_write_args { 2389 struct vnode *a_vp; 2390 struct uio *a_uio; 2391 int a_ioflag; 2392 struct ucred *a_cred; 2393 }; 2394 */ 2395 static int 2396 fuse_vnop_write(struct vop_write_args *ap) 2397 { 2398 struct vnode *vp = ap->a_vp; 2399 struct uio *uio = ap->a_uio; 2400 int ioflag = ap->a_ioflag; 2401 struct ucred *cred = ap->a_cred; 2402 pid_t pid = curthread->td_proc->p_pid; 2403 struct fuse_filehandle *fufh; 2404 int err; 2405 bool closefufh = false, directio; 2406 2407 MPASS(vp->v_type == VREG || vp->v_type == VDIR); 2408 2409 if (fuse_isdeadfs(vp)) { 2410 return ENXIO; 2411 } 2412 2413 if (VTOFUD(vp)->flag & FN_DIRECTIO) { 2414 ioflag |= IO_DIRECT; 2415 } 2416 2417 err = fuse_filehandle_getrw(vp, FWRITE, &fufh, cred, pid); 2418 if (err == EBADF && vnode_mount(vp)->mnt_flag & MNT_EXPORTED) { 2419 /* 2420 * nfsd will do I/O without first doing VOP_OPEN. We 2421 * must implicitly open the file here 2422 */ 2423 err = fuse_filehandle_open(vp, FWRITE, &fufh, curthread, cred); 2424 closefufh = true; 2425 } 2426 if (err) { 2427 SDT_PROBE3(fusefs, , vnops, filehandles_closed, vp, uio, cred); 2428 return err; 2429 } 2430 2431 /* 2432 * Ideally, when the daemon asks for direct io at open time, the 2433 * standard file flag should be set according to this, so that would 2434 * just change the default mode, which later on could be changed via 2435 * fcntl(2). 2436 * But this doesn't work, the O_DIRECT flag gets cleared at some point 2437 * (don't know where). So to make any use of the Fuse direct_io option, 2438 * we hardwire it into the file's private data (similarly to Linux, 2439 * btw.). 2440 */ 2441 directio = (ioflag & IO_DIRECT) || !fsess_opt_datacache(vnode_mount(vp)); 2442 2443 fuse_vnode_update(vp, FN_MTIMECHANGE | FN_CTIMECHANGE); 2444 if (directio) { 2445 off_t start, end, filesize; 2446 bool pages = (ioflag & IO_VMIO) != 0; 2447 2448 SDT_PROBE2(fusefs, , vnops, trace, 1, "direct write of vnode"); 2449 2450 err = fuse_vnode_size(vp, &filesize, cred, curthread); 2451 if (err) 2452 goto out; 2453 2454 start = uio->uio_offset; 2455 end = start + uio->uio_resid; 2456 if (!pages) { 2457 err = fuse_inval_buf_range(vp, filesize, start, 2458 end); 2459 if (err) 2460 goto out; 2461 } 2462 err = fuse_write_directbackend(vp, uio, cred, fufh, 2463 filesize, ioflag, pages); 2464 } else { 2465 SDT_PROBE2(fusefs, , vnops, trace, 1, 2466 "buffered write of vnode"); 2467 if (!fsess_opt_writeback(vnode_mount(vp))) 2468 ioflag |= IO_SYNC; 2469 err = fuse_write_biobackend(vp, uio, cred, fufh, ioflag, pid); 2470 } 2471 fuse_internal_clear_suid_on_write(vp, cred, uio->uio_td); 2472 2473 out: 2474 if (closefufh) 2475 fuse_filehandle_close(vp, fufh, curthread, cred); 2476 2477 return (err); 2478 } 2479 2480 static daddr_t 2481 fuse_gbp_getblkno(struct vnode *vp, vm_ooffset_t off) 2482 { 2483 const int biosize = fuse_iosize(vp); 2484 2485 return (off / biosize); 2486 } 2487 2488 static int 2489 fuse_gbp_getblksz(struct vnode *vp, daddr_t lbn, long *blksz) 2490 { 2491 off_t filesize; 2492 int err; 2493 const int biosize = fuse_iosize(vp); 2494 2495 err = fuse_vnode_size(vp, &filesize, NULL, NULL); 2496 if (err) { 2497 /* This will turn into a SIGBUS */ 2498 return (EIO); 2499 } else if ((off_t)lbn * biosize >= filesize) { 2500 *blksz = 0; 2501 } else if ((off_t)(lbn + 1) * biosize > filesize) { 2502 *blksz = filesize - (off_t)lbn *biosize; 2503 } else { 2504 *blksz = biosize; 2505 } 2506 return (0); 2507 } 2508 2509 /* 2510 struct vnop_getpages_args { 2511 struct vnode *a_vp; 2512 vm_page_t *a_m; 2513 int a_count; 2514 int a_reqpage; 2515 }; 2516 */ 2517 static int 2518 fuse_vnop_getpages(struct vop_getpages_args *ap) 2519 { 2520 struct vnode *vp = ap->a_vp; 2521 2522 if (!fsess_opt_mmap(vnode_mount(vp))) { 2523 SDT_PROBE2(fusefs, , vnops, trace, 1, 2524 "called on non-cacheable vnode??\n"); 2525 return (VM_PAGER_ERROR); 2526 } 2527 2528 return (vfs_bio_getpages(vp, ap->a_m, ap->a_count, ap->a_rbehind, 2529 ap->a_rahead, fuse_gbp_getblkno, fuse_gbp_getblksz)); 2530 } 2531 2532 static const char extattr_namespace_separator = '.'; 2533 2534 /* 2535 struct vop_getextattr_args { 2536 struct vop_generic_args a_gen; 2537 struct vnode *a_vp; 2538 int a_attrnamespace; 2539 const char *a_name; 2540 struct uio *a_uio; 2541 size_t *a_size; 2542 struct ucred *a_cred; 2543 struct thread *a_td; 2544 }; 2545 */ 2546 static int 2547 fuse_vnop_getextattr(struct vop_getextattr_args *ap) 2548 { 2549 struct vnode *vp = ap->a_vp; 2550 struct uio *uio = ap->a_uio; 2551 struct fuse_dispatcher fdi; 2552 struct fuse_getxattr_in *get_xattr_in; 2553 struct fuse_getxattr_out *get_xattr_out; 2554 struct mount *mp = vnode_mount(vp); 2555 struct thread *td = ap->a_td; 2556 struct ucred *cred = ap->a_cred; 2557 char *prefix; 2558 char *attr_str; 2559 size_t len; 2560 int err; 2561 2562 if (fuse_isdeadfs(vp)) 2563 return (ENXIO); 2564 2565 if (fsess_not_impl(mp, FUSE_GETXATTR)) 2566 return EOPNOTSUPP; 2567 2568 err = fuse_extattr_check_cred(vp, ap->a_attrnamespace, cred, td, VREAD); 2569 if (err) 2570 return err; 2571 2572 /* Default to looking for user attributes. */ 2573 if (ap->a_attrnamespace == EXTATTR_NAMESPACE_SYSTEM) 2574 prefix = EXTATTR_NAMESPACE_SYSTEM_STRING; 2575 else 2576 prefix = EXTATTR_NAMESPACE_USER_STRING; 2577 2578 len = strlen(prefix) + sizeof(extattr_namespace_separator) + 2579 strlen(ap->a_name) + 1; 2580 2581 fdisp_init(&fdi, len + sizeof(*get_xattr_in)); 2582 fdisp_make_vp(&fdi, FUSE_GETXATTR, vp, td, cred); 2583 2584 get_xattr_in = fdi.indata; 2585 /* 2586 * Check to see whether we're querying the available size or 2587 * issuing the actual request. If we pass in 0, we get back struct 2588 * fuse_getxattr_out. If we pass in a non-zero size, we get back 2589 * that much data, without the struct fuse_getxattr_out header. 2590 */ 2591 if (uio == NULL) 2592 get_xattr_in->size = 0; 2593 else 2594 get_xattr_in->size = uio->uio_resid; 2595 2596 attr_str = (char *)fdi.indata + sizeof(*get_xattr_in); 2597 snprintf(attr_str, len, "%s%c%s", prefix, extattr_namespace_separator, 2598 ap->a_name); 2599 2600 err = fdisp_wait_answ(&fdi); 2601 if (err != 0) { 2602 if (err == ENOSYS) { 2603 fsess_set_notimpl(mp, FUSE_GETXATTR); 2604 err = EOPNOTSUPP; 2605 } 2606 goto out; 2607 } 2608 2609 get_xattr_out = fdi.answ; 2610 2611 if (ap->a_size != NULL) 2612 *ap->a_size = get_xattr_out->size; 2613 2614 if (uio != NULL) 2615 err = uiomove(fdi.answ, fdi.iosize, uio); 2616 2617 out: 2618 fdisp_destroy(&fdi); 2619 return (err); 2620 } 2621 2622 /* 2623 struct vop_setextattr_args { 2624 struct vop_generic_args a_gen; 2625 struct vnode *a_vp; 2626 int a_attrnamespace; 2627 const char *a_name; 2628 struct uio *a_uio; 2629 struct ucred *a_cred; 2630 struct thread *a_td; 2631 }; 2632 */ 2633 static int 2634 fuse_vnop_setextattr(struct vop_setextattr_args *ap) 2635 { 2636 struct vnode *vp = ap->a_vp; 2637 struct uio *uio = ap->a_uio; 2638 struct fuse_dispatcher fdi; 2639 struct fuse_setxattr_in *set_xattr_in; 2640 struct mount *mp = vnode_mount(vp); 2641 struct thread *td = ap->a_td; 2642 struct ucred *cred = ap->a_cred; 2643 char *prefix; 2644 size_t len; 2645 char *attr_str; 2646 int err; 2647 2648 if (fuse_isdeadfs(vp)) 2649 return (ENXIO); 2650 2651 if (fsess_not_impl(mp, FUSE_SETXATTR)) 2652 return EOPNOTSUPP; 2653 2654 if (vfs_isrdonly(mp)) 2655 return EROFS; 2656 2657 /* Deleting xattrs must use VOP_DELETEEXTATTR instead */ 2658 if (ap->a_uio == NULL) { 2659 /* 2660 * If we got here as fallback from VOP_DELETEEXTATTR, then 2661 * return EOPNOTSUPP. 2662 */ 2663 if (fsess_not_impl(mp, FUSE_REMOVEXATTR)) 2664 return (EOPNOTSUPP); 2665 else 2666 return (EINVAL); 2667 } 2668 2669 err = fuse_extattr_check_cred(vp, ap->a_attrnamespace, cred, td, 2670 VWRITE); 2671 if (err) 2672 return err; 2673 2674 /* Default to looking for user attributes. */ 2675 if (ap->a_attrnamespace == EXTATTR_NAMESPACE_SYSTEM) 2676 prefix = EXTATTR_NAMESPACE_SYSTEM_STRING; 2677 else 2678 prefix = EXTATTR_NAMESPACE_USER_STRING; 2679 2680 len = strlen(prefix) + sizeof(extattr_namespace_separator) + 2681 strlen(ap->a_name) + 1; 2682 2683 fdisp_init(&fdi, len + sizeof(*set_xattr_in) + uio->uio_resid); 2684 fdisp_make_vp(&fdi, FUSE_SETXATTR, vp, td, cred); 2685 2686 set_xattr_in = fdi.indata; 2687 set_xattr_in->size = uio->uio_resid; 2688 2689 attr_str = (char *)fdi.indata + sizeof(*set_xattr_in); 2690 snprintf(attr_str, len, "%s%c%s", prefix, extattr_namespace_separator, 2691 ap->a_name); 2692 2693 err = uiomove((char *)fdi.indata + sizeof(*set_xattr_in) + len, 2694 uio->uio_resid, uio); 2695 if (err != 0) { 2696 goto out; 2697 } 2698 2699 err = fdisp_wait_answ(&fdi); 2700 2701 if (err == ENOSYS) { 2702 fsess_set_notimpl(mp, FUSE_SETXATTR); 2703 err = EOPNOTSUPP; 2704 } 2705 if (err == ERESTART) { 2706 /* Can't restart after calling uiomove */ 2707 err = EINTR; 2708 } 2709 2710 out: 2711 fdisp_destroy(&fdi); 2712 return (err); 2713 } 2714 2715 /* 2716 * The Linux / FUSE extended attribute list is simply a collection of 2717 * NUL-terminated strings. The FreeBSD extended attribute list is a single 2718 * byte length followed by a non-NUL terminated string. So, this allows 2719 * conversion of the Linux / FUSE format to the FreeBSD format in place. 2720 * Linux attribute names are reported with the namespace as a prefix (e.g. 2721 * "user.attribute_name"), but in FreeBSD they are reported without the 2722 * namespace prefix (e.g. "attribute_name"). So, we're going from: 2723 * 2724 * user.attr_name1\0user.attr_name2\0 2725 * 2726 * to: 2727 * 2728 * <num>attr_name1<num>attr_name2 2729 * 2730 * Where "<num>" is a single byte number of characters in the attribute name. 2731 * 2732 * Args: 2733 * prefix - exattr namespace prefix string 2734 * list, list_len - input list with namespace prefixes 2735 * bsd_list, bsd_list_len - output list compatible with bsd vfs 2736 */ 2737 static int 2738 fuse_xattrlist_convert(char *prefix, const char *list, int list_len, 2739 char *bsd_list, int *bsd_list_len) 2740 { 2741 int len, pos, dist_to_next, prefix_len; 2742 2743 pos = 0; 2744 *bsd_list_len = 0; 2745 prefix_len = strlen(prefix); 2746 2747 while (pos < list_len && list[pos] != '\0') { 2748 dist_to_next = strlen(&list[pos]) + 1; 2749 if (bcmp(&list[pos], prefix, prefix_len) == 0 && 2750 list[pos + prefix_len] == extattr_namespace_separator) { 2751 len = dist_to_next - 2752 (prefix_len + sizeof(extattr_namespace_separator)) - 1; 2753 if (len >= EXTATTR_MAXNAMELEN) 2754 return (ENAMETOOLONG); 2755 2756 bsd_list[*bsd_list_len] = len; 2757 memcpy(&bsd_list[*bsd_list_len + 1], 2758 &list[pos + prefix_len + 2759 sizeof(extattr_namespace_separator)], len); 2760 2761 *bsd_list_len += len + 1; 2762 } 2763 2764 pos += dist_to_next; 2765 } 2766 2767 return (0); 2768 } 2769 2770 /* 2771 * List extended attributes 2772 * 2773 * The FUSE_LISTXATTR operation is based on Linux's listxattr(2) syscall, which 2774 * has a number of differences compared to its FreeBSD equivalent, 2775 * extattr_list_file: 2776 * 2777 * - FUSE_LISTXATTR returns all extended attributes across all namespaces, 2778 * whereas listxattr(2) only returns attributes for a single namespace 2779 * - FUSE_LISTXATTR prepends each attribute name with "namespace." 2780 * - If the provided buffer is not large enough to hold the result, 2781 * FUSE_LISTXATTR should return ERANGE, whereas listxattr is expected to 2782 * return as many results as will fit. 2783 */ 2784 /* 2785 struct vop_listextattr_args { 2786 struct vop_generic_args a_gen; 2787 struct vnode *a_vp; 2788 int a_attrnamespace; 2789 struct uio *a_uio; 2790 size_t *a_size; 2791 struct ucred *a_cred; 2792 struct thread *a_td; 2793 }; 2794 */ 2795 static int 2796 fuse_vnop_listextattr(struct vop_listextattr_args *ap) 2797 { 2798 struct vnode *vp = ap->a_vp; 2799 struct uio *uio = ap->a_uio; 2800 struct fuse_dispatcher fdi; 2801 struct fuse_listxattr_in *list_xattr_in; 2802 struct fuse_listxattr_out *list_xattr_out; 2803 struct mount *mp = vnode_mount(vp); 2804 struct thread *td = ap->a_td; 2805 struct ucred *cred = ap->a_cred; 2806 char *prefix; 2807 char *bsd_list = NULL; 2808 char *linux_list; 2809 int bsd_list_len; 2810 int linux_list_len; 2811 int err; 2812 2813 if (fuse_isdeadfs(vp)) 2814 return (ENXIO); 2815 2816 if (fsess_not_impl(mp, FUSE_LISTXATTR)) 2817 return EOPNOTSUPP; 2818 2819 err = fuse_extattr_check_cred(vp, ap->a_attrnamespace, cred, td, VREAD); 2820 if (err) 2821 return err; 2822 2823 /* 2824 * Add space for a NUL and the period separator if enabled. 2825 * Default to looking for user attributes. 2826 */ 2827 if (ap->a_attrnamespace == EXTATTR_NAMESPACE_SYSTEM) 2828 prefix = EXTATTR_NAMESPACE_SYSTEM_STRING; 2829 else 2830 prefix = EXTATTR_NAMESPACE_USER_STRING; 2831 2832 fdisp_init(&fdi, sizeof(*list_xattr_in)); 2833 fdisp_make_vp(&fdi, FUSE_LISTXATTR, vp, td, cred); 2834 2835 /* 2836 * Retrieve Linux / FUSE compatible list size. 2837 */ 2838 list_xattr_in = fdi.indata; 2839 list_xattr_in->size = 0; 2840 2841 err = fdisp_wait_answ(&fdi); 2842 if (err != 0) { 2843 if (err == ENOSYS) { 2844 fsess_set_notimpl(mp, FUSE_LISTXATTR); 2845 err = EOPNOTSUPP; 2846 } 2847 goto out; 2848 } 2849 2850 list_xattr_out = fdi.answ; 2851 linux_list_len = list_xattr_out->size; 2852 if (linux_list_len == 0) { 2853 if (ap->a_size != NULL) 2854 *ap->a_size = linux_list_len; 2855 goto out; 2856 } 2857 2858 /* 2859 * Retrieve Linux / FUSE compatible list values. 2860 */ 2861 fdisp_refresh_vp(&fdi, FUSE_LISTXATTR, vp, td, cred); 2862 list_xattr_in = fdi.indata; 2863 list_xattr_in->size = linux_list_len; 2864 2865 err = fdisp_wait_answ(&fdi); 2866 if (err == ERANGE) { 2867 /* 2868 * Race detected. The attribute list must've grown since the 2869 * first FUSE_LISTXATTR call. Start over. Go all the way back 2870 * to userland so we can process signals, if necessary, before 2871 * restarting. 2872 */ 2873 err = ERESTART; 2874 goto out; 2875 } else if (err != 0) 2876 goto out; 2877 2878 linux_list = fdi.answ; 2879 /* FUSE doesn't allow the server to return more data than requested */ 2880 if (fdi.iosize > linux_list_len) { 2881 struct fuse_data *data = fuse_get_mpdata(mp); 2882 2883 fuse_warn(data, FSESS_WARN_LSEXTATTR_LONG, 2884 "server returned " 2885 "more extended attribute data than requested; " 2886 "should've returned ERANGE instead."); 2887 } else { 2888 /* But returning less data is fine */ 2889 linux_list_len = fdi.iosize; 2890 } 2891 2892 /* 2893 * Retrieve the BSD compatible list values. 2894 * The Linux / FUSE attribute list format isn't the same 2895 * as FreeBSD's format. So we need to transform it into 2896 * FreeBSD's format before giving it to the user. 2897 */ 2898 bsd_list = malloc(linux_list_len, M_TEMP, M_WAITOK); 2899 err = fuse_xattrlist_convert(prefix, linux_list, linux_list_len, 2900 bsd_list, &bsd_list_len); 2901 if (err != 0) 2902 goto out; 2903 2904 if (ap->a_size != NULL) 2905 *ap->a_size = bsd_list_len; 2906 2907 if (uio != NULL) 2908 err = uiomove(bsd_list, bsd_list_len, uio); 2909 2910 out: 2911 free(bsd_list, M_TEMP); 2912 fdisp_destroy(&fdi); 2913 return (err); 2914 } 2915 2916 /* 2917 struct vop_deallocate_args { 2918 struct vop_generic_args a_gen; 2919 struct vnode *a_vp; 2920 off_t *a_offset; 2921 off_t *a_len; 2922 int a_flags; 2923 int a_ioflag; 2924 struct ucred *a_cred; 2925 }; 2926 */ 2927 static int 2928 fuse_vnop_deallocate(struct vop_deallocate_args *ap) 2929 { 2930 struct vnode *vp = ap->a_vp; 2931 struct mount *mp = vnode_mount(vp); 2932 struct fuse_filehandle *fufh; 2933 struct fuse_dispatcher fdi; 2934 struct fuse_fallocate_in *ffi; 2935 struct ucred *cred = ap->a_cred; 2936 pid_t pid = curthread->td_proc->p_pid; 2937 off_t *len = ap->a_len; 2938 off_t *offset = ap->a_offset; 2939 int ioflag = ap->a_ioflag; 2940 off_t filesize; 2941 int err; 2942 bool closefufh = false; 2943 2944 if (fuse_isdeadfs(vp)) 2945 return (ENXIO); 2946 2947 if (vfs_isrdonly(mp)) 2948 return (EROFS); 2949 2950 if (fsess_not_impl(mp, FUSE_FALLOCATE)) 2951 goto fallback; 2952 2953 err = fuse_filehandle_getrw(vp, FWRITE, &fufh, cred, pid); 2954 if (err == EBADF && vnode_mount(vp)->mnt_flag & MNT_EXPORTED) { 2955 /* 2956 * nfsd will do I/O without first doing VOP_OPEN. We 2957 * must implicitly open the file here 2958 */ 2959 err = fuse_filehandle_open(vp, FWRITE, &fufh, curthread, cred); 2960 closefufh = true; 2961 } 2962 if (err) 2963 return (err); 2964 2965 fuse_vnode_update(vp, FN_MTIMECHANGE | FN_CTIMECHANGE); 2966 2967 err = fuse_vnode_size(vp, &filesize, cred, curthread); 2968 if (err) 2969 goto out; 2970 fuse_inval_buf_range(vp, filesize, *offset, *offset + *len); 2971 2972 fdisp_init(&fdi, sizeof(*ffi)); 2973 fdisp_make_vp(&fdi, FUSE_FALLOCATE, vp, curthread, cred); 2974 ffi = fdi.indata; 2975 ffi->fh = fufh->fh_id; 2976 ffi->offset = *offset; 2977 ffi->length = *len; 2978 /* 2979 * FreeBSD's fspacectl is equivalent to Linux's fallocate with 2980 * mode == FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE 2981 */ 2982 ffi->mode = FUSE_FALLOC_FL_PUNCH_HOLE | FUSE_FALLOC_FL_KEEP_SIZE; 2983 err = fdisp_wait_answ(&fdi); 2984 2985 if (err == ENOSYS) { 2986 fsess_set_notimpl(mp, FUSE_FALLOCATE); 2987 goto fallback; 2988 } else if (err == EOPNOTSUPP) { 2989 /* 2990 * The file system server does not support FUSE_FALLOCATE with 2991 * the supplied mode for this particular file. 2992 */ 2993 goto fallback; 2994 } else if (!err) { 2995 /* 2996 * Clip the returned offset to EoF. Do it here rather than 2997 * before FUSE_FALLOCATE just in case the kernel's cached file 2998 * size is out of date. Unfortunately, FUSE does not return 2999 * any information about filesize from that operation. 3000 */ 3001 *offset = MIN(*offset + *len, filesize); 3002 *len = 0; 3003 fuse_vnode_undirty_cached_timestamps(vp, false); 3004 fuse_internal_clear_suid_on_write(vp, cred, curthread); 3005 3006 if (ioflag & IO_SYNC) 3007 err = fuse_internal_fsync(vp, curthread, MNT_WAIT, 3008 false); 3009 } 3010 3011 out: 3012 if (closefufh) 3013 fuse_filehandle_close(vp, fufh, curthread, cred); 3014 3015 return (err); 3016 3017 fallback: 3018 if (closefufh) 3019 fuse_filehandle_close(vp, fufh, curthread, cred); 3020 3021 return (vop_stddeallocate(ap)); 3022 } 3023 3024 /* 3025 struct vop_deleteextattr_args { 3026 struct vop_generic_args a_gen; 3027 struct vnode *a_vp; 3028 int a_attrnamespace; 3029 const char *a_name; 3030 struct ucred *a_cred; 3031 struct thread *a_td; 3032 }; 3033 */ 3034 static int 3035 fuse_vnop_deleteextattr(struct vop_deleteextattr_args *ap) 3036 { 3037 struct vnode *vp = ap->a_vp; 3038 struct fuse_dispatcher fdi; 3039 struct mount *mp = vnode_mount(vp); 3040 struct thread *td = ap->a_td; 3041 struct ucred *cred = ap->a_cred; 3042 char *prefix; 3043 size_t len; 3044 char *attr_str; 3045 int err; 3046 3047 if (fuse_isdeadfs(vp)) 3048 return (ENXIO); 3049 3050 if (fsess_not_impl(mp, FUSE_REMOVEXATTR)) 3051 return EOPNOTSUPP; 3052 3053 if (vfs_isrdonly(mp)) 3054 return EROFS; 3055 3056 err = fuse_extattr_check_cred(vp, ap->a_attrnamespace, cred, td, 3057 VWRITE); 3058 if (err) 3059 return err; 3060 3061 /* Default to looking for user attributes. */ 3062 if (ap->a_attrnamespace == EXTATTR_NAMESPACE_SYSTEM) 3063 prefix = EXTATTR_NAMESPACE_SYSTEM_STRING; 3064 else 3065 prefix = EXTATTR_NAMESPACE_USER_STRING; 3066 3067 len = strlen(prefix) + sizeof(extattr_namespace_separator) + 3068 strlen(ap->a_name) + 1; 3069 3070 fdisp_init(&fdi, len); 3071 fdisp_make_vp(&fdi, FUSE_REMOVEXATTR, vp, td, cred); 3072 3073 attr_str = fdi.indata; 3074 snprintf(attr_str, len, "%s%c%s", prefix, extattr_namespace_separator, 3075 ap->a_name); 3076 3077 err = fdisp_wait_answ(&fdi); 3078 if (err == ENOSYS) { 3079 fsess_set_notimpl(mp, FUSE_REMOVEXATTR); 3080 err = EOPNOTSUPP; 3081 } 3082 3083 fdisp_destroy(&fdi); 3084 return (err); 3085 } 3086 3087 /* 3088 struct vnop_print_args { 3089 struct vnode *a_vp; 3090 }; 3091 */ 3092 static int 3093 fuse_vnop_print(struct vop_print_args *ap) 3094 { 3095 struct fuse_vnode_data *fvdat = VTOFUD(ap->a_vp); 3096 3097 printf("nodeid: %ju, parent nodeid: %ju, nlookup: %ju, flag: %#x\n", 3098 (uintmax_t)VTOILLU(ap->a_vp), (uintmax_t)fvdat->parent_nid, 3099 (uintmax_t)fvdat->nlookup, 3100 fvdat->flag); 3101 3102 return 0; 3103 } 3104 3105 /* 3106 * Get an NFS filehandle for a FUSE file. 3107 * 3108 * This will only work for FUSE file systems that guarantee the uniqueness of 3109 * nodeid:generation, which most don't. 3110 */ 3111 /* 3112 vop_vptofh { 3113 IN struct vnode *a_vp; 3114 IN struct fid *a_fhp; 3115 }; 3116 */ 3117 static int 3118 fuse_vnop_vptofh(struct vop_vptofh_args *ap) 3119 { 3120 struct vnode *vp = ap->a_vp; 3121 struct fuse_vnode_data *fvdat = VTOFUD(vp); 3122 struct fuse_fid *fhp = (struct fuse_fid *)(ap->a_fhp); 3123 _Static_assert(sizeof(struct fuse_fid) <= sizeof(struct fid), 3124 "FUSE fid type is too big"); 3125 struct mount *mp = vnode_mount(vp); 3126 struct fuse_data *data = fuse_get_mpdata(mp); 3127 struct vattr va; 3128 int err; 3129 3130 if (!(data->dataflags & FSESS_EXPORT_SUPPORT)) { 3131 /* NFS requires lookups for "." and ".." */ 3132 SDT_PROBE2(fusefs, , vnops, trace, 1, 3133 "VOP_VPTOFH without FUSE_EXPORT_SUPPORT"); 3134 return EOPNOTSUPP; 3135 } 3136 if ((mp->mnt_flag & MNT_EXPORTED) && 3137 !(data->dataflags & FSESS_NO_OPENDIR_SUPPORT)) 3138 { 3139 /* 3140 * NFS is stateless, so nfsd must reopen a directory on every 3141 * call to VOP_READDIR, passing in the d_off field from the 3142 * final dirent of the previous invocation. But without 3143 * FUSE_NO_OPENDIR_SUPPORT, the FUSE protocol does not 3144 * guarantee that d_off will be valid after a directory is 3145 * closed and reopened. So prohibit exporting FUSE file 3146 * systems that don't set that flag. 3147 * 3148 * But userspace NFS servers don't have this problem. 3149 */ 3150 SDT_PROBE2(fusefs, , vnops, trace, 1, 3151 "VOP_VPTOFH without FUSE_NO_OPENDIR_SUPPORT"); 3152 return EOPNOTSUPP; 3153 } 3154 3155 err = fuse_internal_getattr(vp, &va, curthread->td_ucred, curthread); 3156 if (err) 3157 return err; 3158 3159 /*ip = VTOI(ap->a_vp);*/ 3160 /*ufhp = (struct ufid *)ap->a_fhp;*/ 3161 fhp->len = sizeof(struct fuse_fid); 3162 fhp->nid = fvdat->nid; 3163 if (fvdat->generation <= UINT32_MAX) 3164 fhp->gen = fvdat->generation; 3165 else 3166 return EOVERFLOW; 3167 return (0); 3168 } 3169